An online monitoring and self-adaptive control method for tread rubber extrusion size

Abstract

The application discloses a kind of tread rubber extrusion size on-line monitoring and self-adaptive control method, it is related to tire manufacturing technical field, to solve the technical problem that current tread rubber extrusion size is not easy to monitor and control, the method is as follows: using non-contact multi-source vision sensor array, complete three-dimensional reconstruction, based on the deviation, through adaptive algorithm dynamic adjustment extrusion process in the mouth plate opening, extrusion speed and cooling parameter, realize feedforward-feedback compound closed-loop control, ensure that tread rubber extrusion size is stable, the adaptive algorithm is generated based on fuzzy-PID fusion algorithm.This application realizes the high-frequency synchronous scanning and three-dimensional reconstruction of tread rubber width, thickness and cross-sectional profile by using the non-contact multi-source vision sensor array in step one, solves the problem of lacking prediction model and being unable to predict trends, solves the problem of insufficient data processing and prediction ability of the prior art for tread rubber extrusion.

Description

Technical Field

[0001] This invention relates to the field of tire manufacturing technology, and more specifically, to a method for online monitoring and adaptive control of tread rubber extrusion dimensions. Background Technology

[0002] Tread extrusion is one of the core processes in intelligent tire manufacturing. The stability and uniformity of its extruded dimensions directly determine the tire's driving safety, wear resistance, and service life. Therefore, monitoring and controlling the extruded dimensions of the tread compound is a crucial step in high-quality tire production. Currently, the industry mostly uses contact measurement or single-point non-contact measurement methods to monitor the extruded dimensions of the tread compound. Contact measurement is easily affected by the viscoelasticity of the tread compound at high temperatures, causing measurement errors and damage to the compound surface. Single-point non-contact measurement often uses a single sensor to achieve local dimensional detection, combined with manual monitoring of basic process parameters such as extrusion speed and compound temperature. The control process mainly relies on the operator's experience, manually adjusting parameters such as die opening and extrusion speed. Some companies have introduced conventional fuzzy control or PID control algorithms to assist in control in an attempt to improve dimensional control accuracy.

[0003] Existing technologies have significant limitations and cannot meet the demands of high-quality, efficient, and automated tire production. Current monitoring methods have shortcomings: contact measurements are prone to damaging the rubber compound and lack accuracy; single-point non-contact measurements cannot achieve comprehensive and simultaneous monitoring of tread rubber width, thickness, and cross-sectional profile; and the lack of standardized internal and external parameter calibration for sensors can easily lead to deviations in pixel coordinates versus world coordinates, further affecting measurement accuracy. Furthermore, data processing and prediction capabilities are insufficient, failing to effectively integrate multiple process parameters with monitored dimensional data, lacking high-precision dynamic prediction models, and unable to predict dimensional change trends. Therefore, we propose an online monitoring and adaptive control method for tread rubber extrusion dimensions. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art, adapt to practical needs, and provide an online monitoring and adaptive control method for the extrusion size of tread rubber, so as to solve the current technical problem that the extrusion size of tread rubber is not easy to monitor and control.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an online monitoring and adaptive control method for tread rubber extrusion dimensions, the method being as follows:

[0006] Step 1: Using a non-contact multi-source vision sensor array, the width, thickness, and cross-sectional profile of the high-temperature extruded tread rubber are scanned at high frequency and three-dimensional reconstruction is completed to obtain real-time dimensional data of the tread rubber.

[0007] Step 2: Collect multiple process parameters during the tread rubber extrusion process, and fuse these multiple process parameters with the real-time dimensional data obtained in Step 1;

[0008] Step 3: Based on deep learning algorithms, establish a dynamic prediction model for tread rubber extrusion dimensions using the fused dataset;

[0009] Step 4: Calculate the deviation between the real-time size data obtained in Step 1 and the predicted size data output by the prediction model in Step 3;

[0010] Step 5: Based on the deviation, the die opening, extrusion speed and cooling parameters during the extrusion process are dynamically adjusted through an adaptive algorithm to achieve feedforward-feedback composite closed-loop control, ensuring the stability of the tread rubber extrusion dimensions. The adaptive algorithm is generated based on the fuzzy-PID fusion algorithm.

[0011] Preferably, the non-contact multi-source vision sensor array in step one is composed of a linear array camera and a laser sensor, with a high-frequency synchronous scanning frequency of 30~60fps and a scanning error of less than 1ms.

[0012] Preferably, the 3D reconstruction in step one is processed by combining multi-view parallax reconstruction and 3D point cloud construction, and the specific steps are as follows:

[0013] S1: Using a linear array camera in a non-contact multi-source vision sensor array, the cross-section of the tread rubber extruded at high temperature is simultaneously photographed from at least three different perspectives. The shooting perspectives correspond to the top, left side, and right side of the tread rubber, respectively, to ensure complete coverage of the tread rubber cross-section outline. At the same time, distance data from the laser sensor is collected simultaneously to help eliminate image interference caused by high temperature.

[0014] S2: Denoise, grayscale, and edge enhancement are performed on the acquired multi-view images to remove noise interference caused by high-temperature smoke and light reflection, and to enhance the edge features of the tread rubber cross-section contour.

[0015] S3: Based on the preprocessed multi-view images, a stereo matching algorithm is used to calculate the disparity values ​​of corresponding pixels under different views. Combined with the intrinsic and extrinsic parameter data after sensor calibration, the pixel coordinates are converted into world coordinates to initially construct the spatial position relationship of the two-dimensional contour of the tread rubber section.

[0016] S4: Based on the spatial position data obtained from S3, the distance data collected by the laser sensor is fused together. Through the point cloud generation algorithm, the discrete pixels and distance data are converted into continuous three-dimensional point cloud data, which completely restores the three-dimensional shape of the tread rubber cross section.

[0017] S5: The generated 3D point cloud data is deredundant and smoothed to remove outliers and noise points. Then, the 3D contour of the tread rubber cross section is extracted by the point cloud segmentation algorithm. Finally, the 3D coordinate data of the tread rubber cross section contour, i.e., the size data, is output.

[0018] Preferably, the multiple process parameters in step two include extrusion speed, rubber temperature, rubber pressure, and rubber viscosity. The fusion process adopts a weighted fusion mechanism to eliminate redundant interference between different parameters. Specifically, based on the extrusion process characteristics of the tread rubber, the real-time size data in step one is given a higher basic weight. Among the multiple process parameters, rubber temperature and rubber pressure have a more significant impact on size, and their default weights are higher than those of extrusion speed and rubber viscosity.

[0019] Preferably, the deep learning algorithm in step three is one or more combinations of convolutional neural networks, long short-term memory networks, and gradient boosting regression models. The dynamic prediction model is trained using historical monitoring data, process parameter data, and dimensional deviation data. The evaluation metrics of the dynamic prediction model include the coefficient of determination, root mean square error, and mean absolute error.

[0020] Preferably, the commonality among the convolutional neural network, long short-term memory network, and gradient boosting regression model is that they all possess strong nonlinear fitting capabilities, specifically as follows: During the tread rubber extrusion process, the small changes in the rubber compound, including temperature, pressure, and extrusion speed, have a nonlinear relationship with the changes in tread rubber width and thickness. Furthermore, the viscoelasticity of the rubber compound at high temperatures further exacerbates the nonlinear characteristics. Based on this characteristic, the convolutional neural network achieves nonlinear mapping through local feature extraction of the convolutional kernel and a nonlinear activation function; the long short-term memory network captures temporal nonlinear dependencies through a gating mechanism; and the gradient boosting regression model models complex nonlinear relationships through iterative fitting of the residuals of multiple decision trees.

[0021] Preferably, the specific steps for calculating the data deviation in step four are as follows:

[0022] S1: Align the real-time dimensional data obtained in step one with the corresponding predicted dimensional data output by the prediction model in step three in terms of time and space dimensions to ensure that the two correspond to the same extrusion time and the same cross-sectional position of dimensional parameters;

[0023] S2: Calculate the deviations at each coordinate point of the width, thickness, and cross-sectional profile. The core method combines absolute and relative deviations, and the specific formulas are as follows:

[0024] Width deviation:

[0025] (Absolute deviation) (Relative deviation);

[0026] Thickness deviation:

[0027] (Absolute deviation) (Relative deviation);

[0028] Three-dimensional coordinate deviation of cross-section profile:

[0029] For each three-dimensional coordinate point Yi(x,y,z) on the cross-sectional profile, calculate its Euclidean distance to the corresponding predicted coordinate point Ŷi(x,y,z), i.e. This serves as the dimensional deviation of that coordinate point;

[0030] S3: Comprehensive Deviation Fusion: The above single-dimensional dimensional deviations are weighted and fused to obtain the overall dimensional deviation. ;

[0031] S4: Deviation Check: Verify the calculated overall dimensional deviation. The data is compared with a preset deviation threshold, and abnormal deviation data is eliminated to finally output a valid and accurate dimensional deviation.

[0032] Preferably, the adaptive algorithm in step five is generated based on a fuzzy-PID fusion algorithm. Based on this algorithm, it also includes introducing the size deviation prediction value output by a deep learning prediction model to realize the collaborative logic of predicting deviation trends and real-time correction. Furthermore, it is based on a coupling compensation mechanism designed with multiple process parameters to solve the problem of mutual interference among multiple parameters during the tread rubber extrusion process. The adaptive algorithm includes the following formula:

[0033] Dynamic weight allocation formula:

[0034]

[0035] In the formula, Let k be the PID control weight. Let k be the weights for fuzzy inference. The real-time size deviation at time k. To preset the maximum permissible deviation, Let k be the rate of change of the deviation at time k. To preset the maximum rate of change of deviation, These are weighted adjustment coefficients, and their sum is 1;

[0036] Control command output formula:

[0037]

[0038] In the formula, This is the final control instruction at time k. Let k be the PID control output value at time k. Let k be the fuzzy inference output value. The coupling deviation of multiple process parameters at time k, For coupling compensation coefficients, The deviation of the tread rubber cross-section profile at time k. This represents the contour constraint coefficient.

[0039] Preferably, in step five, the feedforward control of the feedforward-feedback composite closed-loop control specifically involves predicting the trend of size deviation in advance based on a dynamic prediction model and outputting a pre-adjustment command, while the feedback control specifically involves correcting the pre-adjustment command based on the real-time size deviation.

[0040] Preferably, in step one, the non-contact multi-source vision sensor array undergoes calibration, which includes intrinsic parameter calibration and extrinsic parameter calibration to achieve accurate conversion from pixel coordinates to world coordinates.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] 1. This invention utilizes a non-contact multi-source vision sensor array in step one to achieve high-frequency synchronous scanning and three-dimensional reconstruction of tread rubber width, thickness, and cross-sectional contour, replacing contact and single-point measurements. Furthermore, sensor calibration resolves the issue of pixel coordinate and world coordinate conversion deviation, improving monitoring accuracy and addressing the shortcomings of existing detection methods. Simultaneously, step two integrates multiple process parameters with the real-time dimensional data acquired in step one, resolving the problem of data non-fusion. In step three, based on the fused data, a dynamic prediction model is established using deep learning algorithms, addressing the lack of a prediction model and the inability to predict trends. This solves the problem of insufficient processing and prediction capabilities for tread rubber extrusion data in existing technologies.

[0043] 2. In addition to solving the calculation deviation problem in step four, this invention avoids the one-sided control caused by single-dimensional deviation by multi-dimensional deviation calculation and comprehensive weighted fusion. At the same time, it solves the derivative problem of abnormal deviation interfering with control instructions by eliminating abnormal data through deviation verification, thus ensuring the accuracy of control basis. Furthermore, it avoids the misalignment between real-time size and predicted size by aligning data in time and space dimensions, thus solving the derivative problem of deviation calculation distortion caused by data mismatch, which in turn leads to control errors.

[0044] 3. In addition, based on the fuzzy-PID fusion adaptive algorithm in step five, the present invention first obtains the predicted value of size deviation through a deep learning prediction model, integrates multiple process parameters for coupling compensation, then adjusts the weights of fuzzy inference and PID control through a dynamic weight allocation strategy, combines the tread rubber cross-section profile data to constrain the control amplitude, and finally outputs a precise control command to achieve feedforward-feedback composite closed-loop control. Attached Figure Description

[0045] Figure 1This is a schematic diagram of the process structure of the present invention. Detailed Implementation

[0046] like Figure 1 As shown, the present invention relates to an online monitoring and adaptive control method for tread rubber extrusion dimensions, the method being as follows:

[0047] Step 1: Using a non-contact multi-source vision sensor array, the width, thickness, and cross-sectional profile of the high-temperature extruded tread rubber are scanned at high frequency and three-dimensional reconstruction is completed to obtain real-time dimensional data of the tread rubber.

[0048] Step 2: Collect multiple process parameters during the tread rubber extrusion process, and fuse these multiple process parameters with the real-time dimensional data obtained in Step 1.

[0049] Step 3: Based on deep learning algorithms, establish a dynamic prediction model for tread rubber extrusion dimensions using the fused dataset;

[0050] Step 4: Calculate the deviation between the real-time size data obtained in Step 1 and the predicted size data output by the prediction model in Step 3;

[0051] Step 5: Based on the deviation, the die opening, extrusion speed and cooling parameters in the extrusion process are dynamically adjusted through an adaptive algorithm to achieve feedforward-feedback composite closed-loop control, ensuring the stability of the tread rubber extrusion dimensions. The adaptive algorithm is generated based on the fuzzy-PID fusion algorithm.

[0052] This invention utilizes a non-contact multi-source vision sensor array in step one to achieve high-frequency synchronous scanning and three-dimensional reconstruction of tread rubber width, thickness, and cross-sectional contour, replacing contact and single-point measurements. Sensor calibration resolves the issue of pixel coordinate to world coordinate conversion deviation, improving monitoring accuracy and addressing the shortcomings of existing detection methods. Step two integrates multiple process parameters with the real-time dimensional data acquired in step one, resolving the problem of data incompatibility. Step three, based on the fused data, establishes a dynamic prediction model using deep learning algorithms, addressing the lack of a prediction model and the inability to predict trends. This invention solves the problem of insufficient processing and prediction capabilities for tread rubber extrusion data in existing technologies.

[0053] This invention further addresses the issue of computational bias by employing multi-dimensional bias calculation and comprehensive weighted fusion in step four. This avoids the bias caused by single-dimensional bias in regulation. Simultaneously, it eliminates abnormal data through bias verification, resolving the derivative problem of abnormal bias interfering with regulation commands and ensuring the accuracy of regulation basis. Furthermore, it avoids misalignment between real-time and predicted dimensions by aligning data in the time and spatial dimensions, thus resolving the derivative problem of data mismatch leading to distorted bias calculations and subsequent regulation errors.

[0054] Specifically, the non-contact multi-source vision sensor array in step one is composed of a linear scan camera and a laser sensor, with a high-frequency synchronous scanning frequency of 30~60fps and a scanning error of less than 1ms.

[0055] This sensor array combines the visual imaging advantages of a line scan camera with the distance detection advantages of a laser sensor, making it specifically suited for monitoring high-temperature tire tread rubber. The high-frequency scanning of 30~60fps ensures the real-time nature of dimensional data, and the scanning error of less than 1ms avoids data lag. It provides high-precision and timely raw data for subsequent 3D reconstruction and real-time dimensional acquisition, effectively making up for the limitations of traditional single-point measurement.

[0056] More specifically, the 3D reconstruction in step one is handled by combining multi-view parallax reconstruction with 3D point cloud construction. The specific steps are as follows:

[0057] S1: Multi-view image acquisition. Using a line scan camera in a non-contact multi-source vision sensor array, the cross-section of the tread rubber extruded at high temperature is captured simultaneously from at least three different perspectives. The shooting perspectives correspond to the top, left side, and right side of the tread rubber, respectively, to ensure complete coverage of the tread rubber cross-section outline. At the same time, distance data from the laser sensor is acquired simultaneously to help eliminate image interference caused by high temperature.

[0058] S2: Image preprocessing, which involves denoising, grayscale conversion, and edge enhancement of the acquired multi-view images to remove noise interference caused by high-temperature smoke and light reflection, and to enhance the edge features of the tread rubber cross-section contour.

[0059] S3: Multi-view disparity reconstruction. Based on the pre-processed multi-view images, a stereo matching algorithm is used to calculate the disparity values ​​of corresponding pixels under different views. Combined with the intrinsic and extrinsic parameter data after sensor calibration, the pixel coordinates are converted into world coordinates to initially construct the spatial position relationship of the two-dimensional contour of the tread rubber section.

[0060] S4: 3D point cloud construction. Based on the spatial position data obtained from S3 and the distance data collected by the laser sensor, the discrete pixel points and distance data are fused together. Through the point cloud generation algorithm, the discrete pixel points and distance data are converted into continuous 3D point cloud data, which completely restores the 3D shape of the tread rubber cross section.

[0061] S5: Point Cloud Optimization and Contour Extraction. The generated 3D point cloud data undergoes redundancy removal and smoothing to remove outliers and noise. Then, a point cloud segmentation algorithm is used to extract the 3D contour of the tread rubber cross-section, ultimately outputting the 3D coordinate data (i.e., dimensional data) of the tread rubber cross-section contour. Redundancy removal employs a voxel grid downsampling algorithm. By setting a reasonable voxel size, the discrete 3D point cloud is divided into grids, retaining representative points within each voxel, efficiently removing redundant points without destroying the tread rubber cross-section contour features. Smoothing uses the moving least squares (MLS) method. By fitting the surface of the local neighborhood of the point cloud, the point cloud coordinates are corrected, achieving point cloud smoothing while preserving contour details. Outlier and noise removal uses a statistical filtering algorithm. By calculating the mean and standard deviation of the distance between each point and its neighbors, outliers and noise exceeding a preset threshold are eliminated, adapting to the point cloud noise characteristics under high-temperature extrusion scenarios.

[0062] Furthermore, in step two, multiple process parameters include extrusion speed, rubber temperature, rubber pressure, and rubber viscosity. The fusion process adopts a weighted fusion mechanism to eliminate redundant interference between different parameters. Specifically, based on the characteristics of the tread rubber extrusion process, the real-time size data in step one is given a higher basic weight. Moreover, among the multiple process parameters, rubber temperature and rubber pressure have a more significant impact on size, and their default weights are higher than those of extrusion speed and rubber viscosity.

[0063] This weighted fusion mechanism is designed entirely based on the actual process characteristics of tread rubber extrusion. It accurately matches the degree of influence of each parameter on the extrusion size, prioritizes the core role of size data, and distinguishes the weight differences of process parameters. This effectively eliminates redundant interference between different parameters, achieves efficient fusion of monitoring data and process data, and provides a high-quality dataset for the subsequent prediction model.

[0064] Furthermore, in step three, the deep learning algorithm is one or more combinations of convolutional neural networks, long short-term memory networks, and gradient boosting regression models. The dynamic prediction model is trained using historical monitoring data, process parameter data, and dimensional deviation data. The evaluation metrics of this dynamic prediction model include the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error.

[0065] The selected deep learning algorithms all have strong nonlinear fitting capabilities, which can adapt to the complex extrusion process caused by the viscoelasticity of rubber materials. The model is trained with historical monitoring, process and deviation data, which can fit the parameter change law of actual production. Multi-dimensional evaluation indicators verify the accuracy of the model from different perspectives such as fit degree and error value, and comprehensively ensure the reliability and accuracy of the size prediction results.

[0066] It is worth noting that convolutional neural networks, long short-term memory networks, and gradient boosting regression models all share a common characteristic: they all possess strong nonlinear fitting capabilities. Specifically, the relationship between the small changes in the rubber compound during the tread extrusion process, including temperature, pressure, and extrusion speed, and the changes in tread width and thickness is not linear. Furthermore, the viscoelasticity of the rubber compound at high temperatures further exacerbates the nonlinear characteristics. Based on this characteristic, convolutional neural networks achieve nonlinear mapping through local feature extraction of convolutional kernels and nonlinear activation functions; long short-term memory networks capture temporal nonlinear dependencies through gating mechanisms; and gradient boosting regression models model complex nonlinear relationships through iterative fitting of residuals from multiple decision trees.

[0067] The extrusion process parameters and dimensions of tread compound exhibit a complex nonlinear relationship. Under high-temperature conditions, the viscoelasticity of the compound will further exacerbate this characteristic. Three models achieve nonlinear modeling from different dimensions, each with its own technical focus. They can be used individually or in combination, and can flexibly adapt to the production scenarios and data foundations of different tire companies, maximizing the technical advantages of nonlinear fitting.

[0068] It is worth mentioning that the specific steps for calculating the data deviation in step four are as follows:

[0069] S1: Align the real-time dimensional data obtained in step one with the corresponding predicted dimensional data output by the prediction model in step three in terms of time and space dimensions to ensure that the two correspond to the same extrusion time and the same cross-sectional position of dimensional parameters;

[0070] S2: Calculate the deviations at each coordinate point of the width, thickness, and cross-sectional profile. The core method combines absolute and relative deviations, and the specific formulas are as follows:

[0071] Width deviation:

[0072] (Absolute deviation) (Relative deviation);

[0073] Thickness deviation:

[0074] (Absolute deviation) (Relative deviation);

[0075] Three-dimensional coordinate deviation of cross-section profile:

[0076] For each three-dimensional coordinate point Yi(x,y,z) on the cross-sectional profile, calculate its Euclidean distance to the corresponding predicted coordinate point Ŷi(x,y,z), i.e. This serves as the dimensional deviation of that coordinate point;

[0077] S3: Comprehensive Deviation Fusion: The above single-dimensional dimensional deviations are weighted and fused to obtain the overall dimensional deviation. ;

[0078] S4: Deviation Check: Verify the calculated overall dimensional deviation. The deviation is compared with a preset deviation threshold, abnormal deviation data is eliminated, and finally the effective and accurate dimensional deviation is output.

[0079] The deviation calculation process first aligns the spatiotemporal dimensions to fundamentally avoid deviation distortion caused by misalignment between real-time and predicted dimensions. The multi-dimensional deviation calculation takes into account the overall indicators of width and thickness as well as the details of cross-sectional contours. The comprehensive weighted fusion avoids the one-sidedness of single-dimensional control. Deviation verification removes abnormal data, providing accurate and effective deviation basis for subsequent control steps.

[0080] It is worth mentioning that the adaptive algorithm in step five is generated based on the fuzzy-PID fusion algorithm. Based on this algorithm, it also incorporates the size deviation prediction value output by the deep learning prediction model to realize the collaborative logic of predicting deviation trends and real-time correction. Furthermore, it designs a coupling compensation mechanism based on multiple process parameters to solve the problem of mutual interference among multiple parameters during the tread rubber extrusion process. The adaptive algorithm includes the following formula:

[0081] Dynamic weight allocation formula:

[0082]

[0083] In the formula, Let k be the PID control weight. Let k be the weights for fuzzy inference. The real-time size deviation at time k. To preset the maximum permissible deviation, Let k be the rate of change of the deviation at time k. To preset the maximum rate of change of deviation, These are weighted adjustment coefficients, and their sum is 1;

[0084] Control command output formula:

[0085]

[0086] In the formula, This is the final control instruction at time k. Let k be the PID control output value at time k. Let k be the fuzzy inference output value. The coupling deviation of multiple process parameters at time k, For coupling compensation coefficients, The deviation of the tread rubber cross-section profile at time k. The contour constraint coefficient;

[0087] This adaptive algorithm combines the flexibility of fuzzy control with the precision and stability of PID control. The dynamic weight allocation can be adjusted in real time according to the deviation and rate of change. The introduction of predictive values ​​enables trend-based intervention. The coupling compensation mechanism specifically solves the problem of mutual interference among multiple process parameters. The formulaic design makes the output of control commands more quantitative and precise, adapting to the dynamically changing extrusion process.

[0088] It is worth noting that in step five, the feedforward control of the feedforward-feedback composite closed-loop control specifically involves predicting the trend of size deviation in advance based on the dynamic prediction model and outputting pre-adjustment commands, while the feedback control specifically involves correcting the pre-adjustment commands based on the real-time size deviation.

[0089] The present invention also uses step five, based on the fuzzy-PID fusion adaptive algorithm, to first obtain the predicted value of the size deviation through a deep learning prediction model, fuse multiple process parameters for coupling compensation, then adjust the weights of fuzzy inference and PID control through a dynamic weight allocation strategy, combine the tread rubber cross-sectional profile data to constrain the control amplitude, and finally output precise control commands to achieve feedforward-feedback composite closed-loop control.

[0090] This feedforward-feedback composite closed-loop control mode combines the advantages of predictability and real-time performance. The feedforward control predicts the trend of dimensional deviation in advance based on the predictive model and outputs pre-control commands, which greatly reduces the lag of traditional control. The feedback control dynamically corrects the pre-control commands based on real-time measured deviations, effectively making up for prediction errors. The two work together to achieve dynamic control of the extrusion size throughout the entire process.

[0091] It is worth noting that in step one, the non-contact multi-source vision sensor array undergoes calibration. The calibration process includes intrinsic parameter calibration and extrinsic parameter calibration, achieving accurate conversion from pixel coordinates to world coordinates.

[0092] The calibration of the sensor array is a crucial preliminary step for accurate monitoring of the dimensions of high-temperature tire tread rubber. Intrinsic parameter calibration is used to correct the hardware errors of the sensor itself, while extrinsic parameter calibration determines the spatial positional relationship between the sensor and the tire tread rubber being measured. This eliminates the conversion deviation from pixel coordinates to world coordinates at the source, significantly improving the accuracy of subsequent 3D reconstruction and dimensional measurement, and solving the coordinate conversion error problem of traditional measurement.

[0093] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. A method for online monitoring and adaptive control of tread rubber extrusion dimensions, characterized in that, The method is as follows: Step 1: Using a non-contact multi-source vision sensor array, the width, thickness, and cross-sectional profile of the high-temperature extruded tread rubber are scanned at high frequency and three-dimensional reconstruction is completed to obtain real-time dimensional data of the tread rubber. Step 2: Collect multiple process parameters during the tread rubber extrusion process, and fuse these multiple process parameters with the real-time dimensional data obtained in Step 1; Step 3: Based on deep learning algorithms, establish a dynamic prediction model for tread rubber extrusion dimensions using the fused dataset; Step 4: Calculate the deviation between the real-time size data obtained in Step 1 and the predicted size data output by the prediction model in Step 3; Step 5: Based on the deviation, the die opening, extrusion speed and cooling parameters during the extrusion process are dynamically adjusted through an adaptive algorithm to achieve feedforward-feedback composite closed-loop control, ensuring the stability of the tread rubber extrusion dimensions. The adaptive algorithm is generated based on the fuzzy-PID fusion algorithm.

2. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 1, characterized in that, The non-contact multi-source vision sensor array in step one is composed of a linear array camera and a laser sensor. Its high-frequency synchronous scanning frequency is 30~60fps, and the scanning error is less than 1ms.

3. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 1, characterized in that, The 3D reconstruction in step one is processed by combining multi-view parallax reconstruction and 3D point cloud construction. The specific steps are as follows: S1: Using a linear array camera in a non-contact multi-source vision sensor array, the cross-section of the tread rubber extruded at high temperature is simultaneously photographed from at least three different perspectives. The shooting perspectives correspond to the top, left side, and right side of the tread rubber, respectively, to ensure complete coverage of the tread rubber cross-section outline. At the same time, distance data from the laser sensor is collected simultaneously to help eliminate image interference caused by high temperature. S2: Denoise, grayscale, and edge enhancement are performed on the acquired multi-view images to remove noise interference caused by high-temperature smoke and light reflection, and to enhance the edge features of the tread rubber cross-section contour. S3: Based on the preprocessed multi-view images, a stereo matching algorithm is used to calculate the disparity values ​​of corresponding pixels under different views. Combined with the intrinsic and extrinsic parameter data after sensor calibration, the pixel coordinates are converted into world coordinates to initially construct the spatial position relationship of the two-dimensional contour of the tread rubber section. S4: Based on the spatial position data obtained from S3, the distance data collected by the laser sensor is fused together. Through the point cloud generation algorithm, the discrete pixels and distance data are converted into continuous three-dimensional point cloud data, which completely restores the three-dimensional shape of the tread rubber cross section. S5: The generated 3D point cloud data is deredundant and smoothed to remove outliers and noise points. Then, the 3D contour of the tread rubber cross section is extracted by the point cloud segmentation algorithm. Finally, the 3D coordinate data of the tread rubber cross section contour, i.e., the size data, is output.

4. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 1, characterized in that, The multiple process parameters in step two include extrusion speed, rubber temperature, rubber pressure, and rubber viscosity. The fusion process adopts a weighted fusion mechanism to eliminate redundant interference between different parameters. Specifically, based on the extrusion process characteristics of the tread rubber, the real-time size data in step one is given a higher basic weight. Among the multiple process parameters, rubber temperature and rubber pressure have a more significant impact on size, and their default weights are higher than those of extrusion speed and rubber viscosity.

5. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 4, characterized in that, In step three, the deep learning algorithm is one or more combinations of convolutional neural networks, long short-term memory networks, and gradient boosting regression models. The dynamic prediction model is trained using historical monitoring data, process parameter data, and dimensional deviation data. The evaluation metrics of the dynamic prediction model include the coefficient of determination, root mean square error, and mean absolute error.

6. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 5, characterized in that, The commonality among the convolutional neural networks, long short-term memory networks, and gradient boosting regression models is their strong nonlinear fitting ability. Specifically, during the tread rubber extrusion process, the small changes in the rubber compound, including temperature, pressure, and extrusion speed, have a nonlinear relationship with the changes in tread rubber width and thickness. Furthermore, the viscoelasticity of the rubber compound at high temperatures further exacerbates the nonlinear characteristics. Based on this characteristic, the convolutional neural network achieves nonlinear mapping through local feature extraction of the convolutional kernel and nonlinear activation functions; the long short-term memory network captures temporal nonlinear dependencies through a gating mechanism; and the gradient boosting regression model models complex nonlinear relationships through iterative fitting of the residuals of multiple decision trees.

7. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 1, characterized in that, The specific steps for calculating the data deviation in step four are as follows: S1: Align the real-time dimensional data obtained in step one with the corresponding predicted dimensional data output by the prediction model in step three in terms of time and space dimensions to ensure that the two correspond to the same extrusion time and the same cross-sectional position of dimensional parameters; S2: Calculate the deviations at each coordinate point of the width, thickness, and cross-sectional profile. The core method combines absolute and relative deviations, and the specific formulas are as follows: Width deviation: (Absolute deviation) (Relative deviation); Thickness deviation: (Absolute deviation) (Relative deviation); Three-dimensional coordinate deviation of cross-section profile: For each three-dimensional coordinate point Yi(x,y,z) on the cross-sectional profile, calculate its Euclidean distance to the corresponding predicted coordinate point Ŷi(x,y,z), i.e. This serves as the dimensional deviation of that coordinate point; S3: Comprehensive Deviation Fusion: The above single-dimensional dimensional deviations are weighted and fused to obtain the overall dimensional deviation. ; S4: Deviation Check: Verify the calculated overall dimensional deviation. The data is compared with a preset deviation threshold, and abnormal deviation data is eliminated. Finally, the effective and accurate dimensional deviation is output.

8. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 1, characterized in that, In step five, the adaptive algorithm is generated based on a fuzzy-PID fusion algorithm. Based on this algorithm, it also incorporates the size deviation prediction value output by a deep learning prediction model to achieve a collaborative logic of predicting deviation trends and real-time correction. Furthermore, it designs a coupling compensation mechanism based on multiple process parameters to address the mutual interference of multiple parameters during tread rubber extrusion. The adaptive algorithm includes the following formula: Dynamic weight allocation formula: In the formula, Let k be the PID control weight. Let k be the weights for fuzzy inference. The real-time size deviation at time k. To preset the maximum permissible deviation, Let k be the rate of change of the deviation at time k. To preset the maximum rate of change of deviation, These are weighted adjustment coefficients, and their sum is 1; Control command output formula: In the formula, This is the final control instruction at time k. Let k be the PID control output value at time k. Let k be the fuzzy inference output value. The coupling deviation of multiple process parameters at time k. For coupling compensation coefficients, Let k be the deviation of the tread rubber cross-section profile. This represents the contour constraint coefficient.

9. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 8, characterized in that, In step five, the feedforward control of the feedforward-feedback composite closed-loop control specifically involves predicting the trend of size deviation in advance based on a dynamic prediction model and outputting a pre-adjustment command. The feedback control specifically involves correcting the pre-adjustment command based on the real-time size deviation.

10. The method for online monitoring and adaptive control of tread rubber extrusion dimensions according to claim 1, characterized in that, In step one, the non-contact multi-source vision sensor array undergoes calibration, which includes intrinsic parameter calibration and extrinsic parameter calibration, to achieve accurate conversion from pixel coordinates to world coordinates.

Images