Prediction method based on fluororubber glass transition temperature
By constructing a glass transition temperature prediction model based on fluororubber, the problems of high measurement cost and low prediction accuracy in the existing technology are solved, and efficient and accurate Tg prediction is achieved, which meets the requirements of low temperature performance evaluation of FKM under extreme working conditions.
Patent Information
- Application Number
- CN202410914698.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-09
AI Technical Summary
Existing technologies for measuring the glass transition temperature (Tg) of fluororubber are costly and the methods for directly predicting the Tg of materials using basic formulations and process parameters have low accuracy, making it difficult to meet the low-temperature performance evaluation requirements of FKM under extreme conditions.
A glass transition temperature prediction model based on fluororubber was constructed. By collecting process parameters and process formulation data, the data was cleaned, aligned and filtered. Multiple sub-models were constructed using regressors, deep learning algorithms and cluster analysis algorithms. Finally, the glass transition temperature prediction model was determined by weighting.
It enables efficient and accurate prediction of the glass transition temperature of fluororubber, reduces measurement costs, improves prediction accuracy, and meets the low-temperature performance evaluation requirements of FKM under extreme conditions.
Smart Images

Figure CN118942579A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of material parameter measurement, and in particular to a prediction method based on the glass transition temperature of fluororubber. Background Art
[0002] FKM is a type of polymer elastomer made of fluorinated olefins. Its unique comprehensive properties such as high temperature resistance, oil resistance, aging resistance, and acid and alkali resistance make it play an indispensable role in aerospace, electronic devices, and chemical industries. However, due to its own chemical structure, FKM has poor low-temperature resistance. In practical applications, the brittle temperature of ordinary FKM is above -30°C, which greatly limits its use in extreme working conditions. In the face of new requirements put forward by application scenarios, it is urgent to improve the low-temperature performance of FKM. In the process of developing and researching FKM, it is crucial to accurately evaluate its low-temperature resistance. Traditionally, the evaluation of such properties relies on multiple test parameters, including brittle temperature, compression cold resistance coefficient, low-temperature shrinkage temperature, and Tg. Especially in the field of aerospace, early studies mostly used brittle temperature and compression cold resistance coefficient to characterize the low-temperature resistance of materials. However, the determination of these parameters involves mixing FKM raw rubber with materials such as additives and carbon black, and vulcanizing it to prepare vulcanized rubber. This process is not only complicated, but also labor- and time-intensive. Although Tr10 can provide information about the performance of materials in low-temperature environments, the method has a long test cycle and high energy consumption, which is particularly inconvenient in research with limited resources and time. In contrast, the measurement of Tg provides a more efficient option. Tg can be directly measured from FKM raw rubber (a white to pale yellow solid) after the polymerization process, greatly reducing energy consumption and reducing operational complexity. Because Tg measurement can save valuable time and resources, and is easy to operate and cost-effective, it has become the preferred indicator for evaluating the low-temperature resistance of FKM. In addition, Tg is not only an important indicator for measuring the application performance of rubber materials in low-temperature environments, but also directly determines the practical temperature range of rubber. Therefore, during the development process, accurate measurement and control of Tg is crucial to ensure that FKM meets actual application needs.
[0003] Tg is a key parameter of rubber materials, marking the critical point where rubber changes from a hard and brittle glassy state to a soft and elastic rubbery state. When the temperature is lower than Tg, the rubber loses its elasticity, becomes hardened and cannot be used effectively. This transition temperature is not only the critical point where the movement of rubber segments starts or stops, but is also significantly affected by environmental conditions. Commonly used methods for measuring Tg include differential scanning calorimetry (DSC), dynamic mechanical analysis (DMA) and static thermomechanical analysis (TMA), which can characterize the thermal, mechanical and viscoelastic properties of materials. However, these techniques rely on expensive equipment and a large number of samples, with high research and development and time costs. There are only ordinary equipment in the laboratory and the sample volume is small, so it is difficult to use such methods.
[0004] At the same time, many aspects of research have been carried out in combining machine learning models to predict Tg. Existing research has proven the effectiveness of machine learning in predicting the Tg of materials, and has achieved remarkable results in analyzing the chemical composition, molecular topological structure and chemical descriptors of formula monomers. However, in the process of material development, precise control of formula design and process parameters is still required to truly optimize material performance. At present, there are still few studies on directly predicting the Tg of materials using basic formulas and process parameters, and the prediction accuracy is low. Summary of the invention
[0005] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a prediction method based on the glass transition temperature of fluororubber. The present invention solves the problems in the prior art of high cost of measuring Tg and limited method of directly predicting material Tg using basic formula and process parameters, resulting in low prediction accuracy.
[0006] To achieve the above object, the present invention provides the following solutions:
[0007] A prediction method based on the glass transition temperature of fluororubber, comprising:
[0008] Obtain the fluororubber data to be tested;
[0009] Inputting the fluororubber data to be tested into the final glass transition temperature prediction model to obtain a transition temperature prediction result;
[0010] The final glass transition temperature prediction model is constructed as follows:
[0011] Collecting fluororubber experimental data, the fluororubber experimental data including: process parameters and process formula;
[0012] Performing data cleaning on the fluororubber experimental data to obtain cleaned data;
[0013] Based on a comprehensive analysis method, data alignment is performed on the cleaned data to determine the defining index of each process parameter and a characteristic sequence of the cleaned data is determined according to the defining index;
[0014] Screening the characteristic sequence to obtain a screened characteristic sequence;
[0015] Respectively constructing a first glass transition temperature prediction sub-model, a second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model based on a regressor, a deep learning algorithm and a cluster analysis algorithm and according to the screened feature sequence;
[0016] A final glass transition temperature prediction model is determined according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model.
[0017] Preferably, the process formula comprises:
[0018] Monomers and auxiliary agents, wherein the monomers include tetrafluoroethylene, perfluoromethyl vinyl ether, hexafluoropropylene and vinylidene fluoride.
[0019] Preferably, the process parameters include:
[0020] Temperature, pressure, feed amount and reaction medium.
[0021] Preferably, the fluororubber experimental data is cleaned to obtain cleaned data, including:
[0022] Determine the monomer content value not included in the process recipe and fill the monomer content value to zero to obtain first sub-cleaning data;
[0023] Determine missing values in the process parameters that cannot be aligned with the original recipe or final performance and delete the missing values to obtain second sub-cleaning data;
[0024] Cleaned data is obtained according to the first sub-cleaning data and the second sub-cleaning data.
[0025] Preferably, the method of aligning the cleaned data based on the comprehensive analysis method, determining the defining index of each process parameter and determining the characteristic sequence of the cleaned data according to the defining index includes:
[0026] Determine a matrix sequence of the cleaned data;
[0027] Performing process parameter estimation on the matrix sequence to obtain a process sequence;
[0028] Obtaining a recombinant sequence according to the process sequence;
[0029] The characteristic sequence of the cleaned data is determined according to the recombinant sequence and the defining index.
[0030] Preferably, the screening of the characteristic sequence to obtain the screened characteristic sequence comprises:
[0031] Determine data correlation according to the Pearson correlation coefficient calculation formula, wherein the data correlation includes: correlation coefficient values between features and performance and direct correlation coefficient values between different features;
[0032] Characterizing the data correlation according to a heat map to obtain a correlation characterization heat map;
[0033] A filtered feature sequence is obtained according to the correlation characterization heat map and the recursive feature elimination algorithm.
[0034] Preferably, determining the final glass transition temperature prediction model according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model comprises:
[0035] Determine a first transition temperature prediction sub-result, a second transition temperature prediction sub-result, and a third transition temperature prediction sub-result respectively according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model, and the third glass transition temperature prediction sub-model;
[0036] Respectively evaluating the first transition temperature prediction sub-result, the second transition temperature prediction sub-result, and the third transition temperature prediction sub-result to obtain a first evaluation result, a second evaluation result, and a third evaluation result;
[0037] The weights of the respective prediction sub-models are determined according to the first evaluation result, the second evaluation result and the third evaluation result, and the weights are added to determine a final glass transition temperature prediction model.
[0038] The present invention discloses the following technical effects:
[0039] The present invention provides a prediction method based on the glass transition temperature of fluororubber, comprising: obtaining fluororubber data to be tested; inputting the fluororubber data to be tested into a final glass transition temperature prediction model to obtain a transition temperature prediction result; the final glass transition temperature prediction model is constructed by: collecting fluororubber experimental data, the fluororubber experimental data comprising: process parameters and process formulas; performing data cleaning on the fluororubber experimental data to obtain cleaned data; based on a comprehensive analysis method, performing data alignment on the cleaned data to determine the defining indicators of each process parameter and determining a feature sequence of the cleaned data according to the defining indicators; screening the feature sequence to obtain a screened feature sequence; constructing a first glass transition temperature prediction sub-model, a second glass transition temperature prediction sub-model and a third glass transition temperature prediction sub-model based on a regressor, a deep learning algorithm and a cluster analysis algorithm respectively and according to the screened feature sequence; and determining a final glass transition temperature prediction model according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model. The present invention takes into account the formulation design and process parameters to directly predict the Tg of the material and uses a variety of models to predict the results, which can more widely target a variety of types of input data to ensure the best prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0041] Figure 1 A flow chart of a prediction method based on the glass transition temperature of fluororubber provided in an embodiment of the present invention;
[0042] Figure 2 A schematic diagram of experimental data categories provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0045] like Figure 1 As shown, the present invention provides a prediction method based on the glass transition temperature of fluororubber, comprising:
[0046] Step 100: Acquire the fluororubber data to be tested;
[0047] Step 200: inputting the fluororubber data to be tested into the final glass transition temperature prediction model to obtain a transition temperature prediction result;
[0048] The final glass transition temperature prediction model is constructed as follows:
[0049] Step 201: collecting fluororubber experimental data, wherein the fluororubber experimental data includes: process parameters and process formula;
[0050] Step 202: Cleaning the fluororubber experimental data to obtain cleaned data;
[0051] Step 203: Based on a comprehensive analysis method, align the cleaned data, determine the defining index of each process parameter, and determine the characteristic sequence of the cleaned data according to the defining index;
[0052] Step 204: screening the characteristic sequence to obtain a screened characteristic sequence;
[0053] Step 205: constructing a first glass transition temperature prediction sub-model, a second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model based on the regressor, the deep learning algorithm and the cluster analysis algorithm and according to the screened feature sequence respectively;
[0054] Step 206: Determine a final glass transition temperature prediction model according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model.
[0055] Furthermore, the process formula includes:
[0056] Monomers and additives, wherein the monomers include tetrafluoroethylene, perfluoromethyl vinyl ether, hexafluoropropylene and vinylidene fluoride. The process parameters include temperature, pressure, feed amount and reaction medium.
[0057] Specifically, the present invention systematically organizes the data from the actual production experiment, including 613 experimental record data, which details the experimental values of the formula composition monomers, processes and performance. According to the confidentiality agreement, the data is encoded with alphanumeric characters. It contains 4 formula monomers and 14 process parameters. Each record represents the material type used to produce FKM, the process record parameters, and the physical and chemical properties of the product produced under the formula composition monomers and process control conditions. Figure 2 This is the FKM data collected in the test, in which the formula monomers mainly include tetrafluoroethylene (TFE), perfluoromethyl vinyl ether (PMVE), hexafluoropropylene (HFP), and vinylidene fluoride (VDF), four fluorinated olefins. The process parameters include temperature (reactor gas phase temperature, reactor liquid phase temperature, reactant feed temperature), pressure (reactor pressure, reactant feed pressure), feed amount, reaction medium and additives. The performance data includes Mooney viscosity and Tg. The original ratio range of the four monomers fluctuates little. At the same time, HFP is not used in the basic formula. The maximum value of Tg is -17℃ and the minimum value is -40℃.
[0058] Furthermore, the fluororubber experimental data is cleaned to obtain cleaned data, including:
[0059] Determine the monomer content value not included in the process recipe and fill the monomer content value to zero to obtain first sub-cleaning data;
[0060] Determine missing values in the process parameters that cannot be aligned with the original recipe or final performance and delete the missing values to obtain second sub-cleaning data;
[0061] Cleaned data is obtained according to the first sub-cleaning data and the second sub-cleaning data.
[0062] Specifically, the original data set was collected by manual records. In order to ensure the quality of the data and the final prediction effect, this study adopted a strategy of deleting or filling the missing values in the formula, process and performance data. For the formula data, in order to maintain the consistency of the combined formula and the integrity of the characteristic dimension, we chose to fill the monomer content values not included in the original formula with zero. This can ensure that the data of various formulas have the same characteristic structure. As for the process data, we noticed that only a small amount of data could not be aligned with the original formula or the final performance results. Taking into account the prediction effect of the model, these missing values were deleted to avoid inaccurate data caused by filling. We evaluated the overall distribution and fluctuation of the data through the parallel coordinate system. Under the guidance of domain experts, outliers were accurately identified. For these erroneous values, replacement and correction measures were taken. For those outliers that exceeded the reasonable fluctuation range, they were deleted to ensure that they would not affect the training and prediction accuracy of the model. In the development of FKM, in addition to the key monomer ratio, process parameters such as reactor pressure and temperature are also crucial. Due to the confidentiality of the test data, the collected process parameters lack continuous time series information, and the process parameters collected in each process stage are of different categories, resulting in data imbalance and the inability to determine the key defining indicators of the parameters. The data imbalance problem here refers to the non-fixed number of records and the non-uniform categories of the recorded process parameters. The inability to determine the parameter defining indicators refers to the fact that not every recorded data has an impact on the final result, so the specific impact of each data cannot be determined. Both of the above problems may lead to inaccurate predictions of the final value. Therefore, this study adopts a comprehensive analysis method, considering the dynamic changes of temperature and pressure during the entire preparation process, and aligning the data after cleaning the data to solve the above problems. In-depth exchanges were conducted with the FKM development team, and based on their professional experience, the fluctuations of the reactor liquid phase temperature, reactant feed temperature, reactor pressure, and reactant feed pressure during the experiment were analyzed in detail, and the same process parameters of this type in each test stage, as well as the key parameters recommended by the researchers, were extracted, and the statistics of each parameter were calculated as the defining indicators of each batch of data. This process is called alignment.
[0063] Furthermore, the method based on comprehensive analysis performs data alignment on the cleaned data, determines the defining index of each process parameter and determines the characteristic sequence of the cleaned data according to the defining index, including:
[0064] Determine a matrix sequence of the cleaned data;
[0065] Performing process parameter estimation on the matrix sequence to obtain a process sequence;
[0066] Obtaining a recombinant sequence according to the process sequence;
[0067] The characteristic sequence of the cleaned data is determined according to the recombinant sequence and the defining index.
[0068] Specifically, the original process data is taken as a matrix sequence S = {C1, C2, C3, L, Cn}, where the matrix C can be expressed as the formula;
[0069]
[0070] Among them, Ti-Pi represents various process parameters recorded in each experiment, Ti represents the liquid phase temperature of the reactor, and Pi represents the pressure in the reactor. The process parameter estimates (minimum, maximum, quartile, mean, variance, mode) of the matrix are extracted to obtain the process sequence E = {Ti_min, Ti_max, Ti_q1, Ti_q2, L, Pi_q3, Pi_mean, Pi_std, Pi_href}. After alignment, there are 32 process estimates. After encoding the estimate names, the recombinant sequence F = {D1, D2, D3, L, Dn} is obtained, where D can be expressed as the formula D = {f7, f8, f9, L, f38};
[0071] The recombinant sequence F was preliminarily determined as the process characteristic parameter for prediction. The process data was matched with the formula ratio according to the product batch, and the characteristic sequence X = {f1, f2, f3, L, f38} was obtained, which included product type (f1), product batch number (f2), 4 monomers (f3-f6), 8 reactor liquid phase temperatures (f7-f14), 8 reactor pressures (f15-f22), 8 reactant feed temperatures (f23-f30), and 8 reactant feed pressures (f31-f38).
[0072] Furthermore, the screening of the characteristic sequence to obtain the screened characteristic sequence includes:
[0073] Determine data correlation according to the Pearson correlation coefficient calculation formula, wherein the data correlation includes: correlation coefficient values between features and performance and direct correlation coefficient values between different features;
[0074] Characterizing the data correlation according to a heat map to obtain a correlation characterization heat map;
[0075] A filtered feature sequence is obtained according to the correlation characterization heat map and the recursive feature elimination algorithm.
[0076] Specifically, the diversity of process parameters leads to the selection of multiple combinations of features, and the reasonable selection depends largely on the researcher's domain knowledge, experience, and prudence. In the experiment, not all process parameters are important for Tg prediction. Irrelevant parameters often lead to an increase in the dimension of input features, thereby reducing the predictive ability of the model. Therefore, it is crucial to screen the best subset of features. In order to enhance the dimension of features and reduce overfitting, the statistical characteristics of process parameters are combined, and alignment is done in the data preprocessing process. Subsequently, the correlation coefficient method is used to further refine the feature selection process. Since the original process parameter data is non-continuous data, we calculate it according to the Pearson correlation coefficient formula:
[0077]
[0078] Calculate the correlation between each sequence feature and the final test result Tg. Where X is any process parameter, i.e., input feature variable, Y is any prediction performance, i.e., target variable, E(X) is the mean of any process parameter feature column, and the calculation formula is as follows. Similarly, E(Y) is the mean of the target variable;
[0079]
[0080] Another σ X is the variance of any attribute feature column, and the calculation formula is as follows. Similarly, σ Y is the variance of the target variable:
[0081]
[0082] Finally, calculate the covariance of any attribute feature and the target variable, formula:
[0083]
[0084] Finally, replace all the formulas into the original formula to get the following formula:
[0085]
[0086] The characteristic elements with strong correlation are eliminated through the correlation coefficient value, and the key features that affect Tg are retained. In order to make the data more intuitive, the correlation coefficient is characterized in combination with the heat map. We preliminarily screen out the features for modeling through the correlation coefficient values between features and the coefficient values between features and performance. In this way, some estimates of f1 (product type), monomers (f3, f4, f5, f6), reactor liquid phase temperature, reactant feed temperature, reactor pressure and reactant feed pressure are selected as input features of the model. At the same time, in the process of model training, in order to realize the automatic selection of features, we introduced feature importance analysis in order to adopt the recursive feature elimination algorithm (RFE), through iterative selection of attributes with strong feature importance, recursively delete features with low contribution at each stage, until the required number of features are obtained.
[0087] Furthermore, in production trials, a three-stage engineering technology system is usually adopted, including small-scale trials, medium-scale trials, and engineering-scale pilot production stages. Although these three stages involve the production of the same type of FKM products, there are significant differences in the monomer ratios in the formula and the records of process parameters. In order to fairly consider all variables and avoid model overfitting, the data partitioning technology in the field of machine learning is used to use 80% of the data as a training set to train and optimize the model, and the remaining 20% as a test set to prepare for further training and verification of the model;
[0088]
[0089] Further, determining a final glass transition temperature prediction model according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model comprises:
[0090] Determine a first transition temperature prediction sub-result, a second transition temperature prediction sub-result, and a third transition temperature prediction sub-result respectively according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model, and the third glass transition temperature prediction sub-model;
[0091] Respectively evaluating the first transition temperature prediction sub-result, the second transition temperature prediction sub-result, and the third transition temperature prediction sub-result to obtain a first evaluation result, a second evaluation result, and a third evaluation result;
[0092] The weights of the respective prediction sub-models are determined according to the first evaluation result, the second evaluation result and the third evaluation result, and the weights are added to determine a final glass transition temperature prediction model.
[0093] Specifically, the error between each prediction sub-result and the actual observed value is calculated to obtain the first evaluation result, the second evaluation result and the third evaluation result; using the first, second and third evaluation results, various methods can be used to determine the weights of each prediction sub-model, such as weighting the weights according to factors such as error size and accuracy; a common method is to use regression analysis or machine learning algorithms to determine the weights of each prediction sub-model so that the error between the prediction result of the final model and the actual observed value is minimized; the results of each prediction sub-model are weighted according to the corresponding weights to obtain the results of the final glass transition temperature prediction model; the final prediction result = weight 1 * first transition temperature prediction sub-result + weight 2 * second transition temperature prediction sub-result + weight 3 * third transition temperature prediction sub-result. For individual special data, the weight can be 0, and the final glass transition temperature prediction model may be a combination of a single or multiple sub-models.
[0094] Furthermore, after feature screening, the entire data set has four monomers, 15 process parameters and product type as input feature variables, with Tg as the output variable.
[0095] Furthermore, in addition to using machine learning models to accurately predict Tg, the ultimate expectation in both experimental research and production processes is to explore the key features that affect the Tg value. Therefore, the present invention designs a visualization framework for interpreting model results and exploring production processes. Focusing on feature importance and attribute feature space analysis, and model prediction process transparent analysis, this meticulous visualization strategy not only reveals the inherent mechanisms of different prediction models, but also greatly enhances the transparency and credibility of the model, providing intuitive visualization support for the formula-process parameters-performance association of FKM materials.
[0096] In order to improve the transparency and understandability of FKM performance prediction analysis and strengthen the basis for decision-making, feature importance analysis is particularly critical. In the process of building a prediction model, each feature is regarded as a potential "driving factor" of the target variable. The feature importance value is a quantitative score of the influence of each feature in the sample, which directly reflects the contribution of the feature to the prediction of the target variable. In order to deeply explain the predictive ability of the model, we not only consider the direct effects between features, but also analyze the interactions within and between feature groups. To this end, we use a network heat map tool (such as https: / / www.omicshare.com) to explore these complex relationships in a visual way. On the one hand, the correlation between features is revealed by calculating the Pearson correlation coefficient, and on the other hand, the importance of features is comprehensively evaluated by quantifying the contribution of features to the model. In addition, the tool can also reveal the internal connection and cross-effect between the prediction results and each feature;
[0097] Combine heat maps and network diagrams to deeply analyze the interaction between features and their impact on prediction results. The intra-group correlation heat map in the upper right corner shows the Pearson correlation coefficient calculated during the feature selection process. In this heat map, the horizontal and vertical axes correspond to different features, and the color block in each cell encodes the positive and negative correlation coefficient. The size of the color block is proportional to the absolute value of the correlation coefficient.
[0098] In addition, the network connection diagram further reveals the correlation between different features and the prediction results, where the connecting lines link the prediction performance to each feature, the line thickness indicates the strength of the correlation, and the lines of different colors are used to distinguish the significance level of the feature importance.
[0099] The correlation of each feature in the prediction of Tg by the ENR model and its contribution to the model prediction. In this network diagram, the three purple lines clearly indicate the decisive role played by the feature product type (f1), monomer VDF and PMVE (f3 and f5) in the prediction model. The remaining light blue lines also indicate that the Tg value has the lowest correlation and contribution with the monomer TFE (f4). At the same time, the heat map highlights the collection of features with strong correlations between features, several estimates of the reactor liquid temperature (f7, f9, f10, f11) and some estimates of the reactant feed temperature (f26, f27, f28, f29), which means that some of them will be given priority when eliminating feature gradients, and individual estimates will be retained. In the research and development process, the attention paid to the process parameter estimates can be appropriately reduced.
[0100] To better reveal the detailed changes during the experiment, we conducted an in-depth visual analysis of the two best-performing prediction models. We first analyzed the RFR model, which performs regression analysis by integrating the prediction results of multiple decision trees. To gain a deeper understanding of the decision-making mechanism of the RFR model, we assigned weights to the best-performing decision trees and selected the subtree with the smallest error for further analysis using the mean square error. We visualized the decision path of the subtree, clearly showing the segmentation of the input features at each node, the distribution of samples at the leaf nodes, and the decision logic of the prediction process.
[0101] Under the maximum tree depth limit, how does a single subtree perform regression prediction on the Tg of FKM material? The path from the root node to the leaf node of the tree is determined by the characteristic test value of each node. The value of each leaf node represents the average value of the Tg of all training instances under the node, representing the prediction results of these instances. In the figure, the decision paths of the three instances on the optimal subtree are marked with different colors. Although the final prediction result of each instance is determined by the mean of the prediction results of all subtrees, the key features of the model can be identified through these paths, which is very helpful for subsequent visual analysis and judgment.
[0102] For models without tree structures, we use SHAP values. SHAP (Shapley Additive ex Planations) is an additive feature attribution model interpretation method based on cooperative game theory. It can not only measure the importance of features, but also provide local and global interpretations of the model. It uses color-coded heat maps, combined with line charts and bar charts to comprehensively display the positive and negative impacts of each feature attribute on the model and its association with the model output. The heat map matrix shows the prediction results of the model, while the bar chart intuitively shows the global importance of each input feature. The horizontal axis of the figure represents the number of instances, and the vertical axis lists different features in order of overall importance. This method identifies product type, VDF, TFE, PMVE (f1, f4, f5, f3) as the key influencing features in the random forest regression model, highlighting the importance of these features in the prediction process. Compared with the monomer features, the influence of the reactor liquid phase temperature (f10, f13, f7) is lower. We observed that monomers (f5, f3) and reactor liquid temperature f10 contributed most to the ENR model, while other features had insignificant effects.
[0103] In order to enhance the local explanatory power of the model, the SHAP values of the RFR and ENR models were used to analyze individual examples, highlighting the specific impact of different features. Taking three examples as examples, we observed that in Example 1, when the Tg value fluctuated around -23°C, the characteristic product type f1, formula monomer (f4, f5, f6) and reactor liquid phase temperature (f9, f12) had a positive impact on the prediction results, and according to the width of the red bar, the influence of product type (f1) and formula monomer (f5, f4) was more significant. Among them, monomer f3 played a negative role. In the ENR model, monomers (f3, f5) and reactor liquid phase temperature f10 and reactant feed pressure (f31, f32) had an impact on the final result. In Example 3, when the Tg value fluctuated around -31°C, product type f1, monomers (f4, f5, f6) and reactor liquid phase temperature (f9, f12) showed a negative impact, which played a decisive role in the prediction value. In the ENR model, characteristic monomers (f3, f5), reactor liquid temperature f10 and reactant feed pressure (f30, f31) also play a negative role.
[0104] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0105] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A prediction method based on the glass transition temperature of fluororubber, It is characterized in that include: Obtain the fluororubber data to be tested; Inputting the fluororubber data to be tested into the final glass transition temperature prediction model to obtain a transition temperature prediction result; The final glass transition temperature prediction model is constructed as follows: Collecting fluororubber experimental data, the fluororubber experimental data including: process parameters and process formula; Performing data cleaning on the fluororubber experimental data to obtain cleaned data; Based on a comprehensive analysis method, data alignment is performed on the cleaned data to determine the defining index of each process parameter and a characteristic sequence of the cleaned data is determined according to the defining index; Screening the characteristic sequence to obtain a screened characteristic sequence; Respectively constructing a first glass transition temperature prediction sub-model, a second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model based on a regressor, a deep learning algorithm and a cluster analysis algorithm and according to the screened feature sequence; A final glass transition temperature prediction model is determined according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model.
2. A prediction method based on the glass transition temperature of fluororubber according to claim 1, It is characterized in that The process formula comprises: Monomers and auxiliary agents, wherein the monomers include tetrafluoroethylene, perfluoromethyl vinyl ether, hexafluoropropylene and vinylidene fluoride.
3. The prediction method based on the glass transition temperature of fluororubber according to claim 1, It is characterized in that The process parameters include: Temperature, pressure, feed amount and reaction medium.
4. The prediction method based on the glass transition temperature of fluororubber according to claim 1, It is characterized in that The fluororubber experimental data is cleaned to obtain cleaned data, including: Determine the monomer content value not included in the process recipe and fill the monomer content value to zero to obtain first sub-cleaning data; Determine missing values in the process parameters that cannot be aligned with the original recipe or final performance and delete the missing values to obtain second sub-cleaning data; Cleaned data is obtained according to the first sub-cleaning data and the second sub-cleaning data.
5. The prediction method based on the glass transition temperature of fluororubber according to claim 1, It is characterized in that The method of aligning the cleaned data based on the comprehensive analysis method, determining the defining index of each process parameter and determining the characteristic sequence of the cleaned data according to the defining index, includes: Determine a matrix sequence of the cleaned data; Performing process parameter estimation on the matrix sequence to obtain a process sequence; Obtaining a recombinant sequence according to the process sequence; The characteristic sequence of the cleaned data is determined according to the recombinant sequence and the defining index.
6. The prediction method based on the glass transition temperature of fluororubber according to claim 1, It is characterized in that The step of screening the characteristic sequence to obtain a screened characteristic sequence includes: Determine data correlation according to the Pearson correlation coefficient calculation formula, wherein the data correlation includes: correlation coefficient values between features and performance and direct correlation coefficient values between different features; Characterizing the data correlation according to a heat map to obtain a correlation characterization heat map; A filtered feature sequence is obtained according to the correlation characterization heat map and the recursive feature elimination algorithm.
7. The prediction method based on the glass transition temperature of fluororubber according to claim 1, It is characterized in that Determining a final glass transition temperature prediction model according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model and the third glass transition temperature prediction sub-model comprises: Determine a first transition temperature prediction sub-result, a second transition temperature prediction sub-result, and a third transition temperature prediction sub-result respectively according to the first glass transition temperature prediction sub-model, the second glass transition temperature prediction sub-model, and the third glass transition temperature prediction sub-model; Respectively evaluating the first transition temperature prediction sub-result, the second transition temperature prediction sub-result, and the third transition temperature prediction sub-result to obtain a first evaluation result, a second evaluation result, and a third evaluation result; The weights of the respective prediction sub-models are determined according to the first evaluation result, the second evaluation result and the third evaluation result, and the weights are added to determine a final glass transition temperature prediction model.
Citation Information
Patent Citations
System and method for predicting quality of a chemical compound and / or of a formulation thereof as a product of a production process
CA3112860A1
Method and system for predicting glass transition temperature of polyimide film
CN117059211A
Method for detecting tannin content in wine brewing grains
CN117288692A
Oxide glass performance prediction method based on interpretable high-dimensional space prediction model
CN117497087A
Polyimide glass transition temperature prediction method based on interpretable machine learning
CN118053514A
