A system and method for validation of a material formulation by confidence intervals

Abstract

The application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision-making technical field.The present application discloses a kind of verification system and method to material formula by confidence interval, it is related to material informatics and intelligent research and development decision

Description

TECHNICAL FIELD

[0001] The present application relates to the field of material informatics and intelligent research and development decision-making, in particular to a system and method for verifying material formula through confidence interval. BACKGROUND

[0002] In the field of material research and development, the prediction and evaluation equipment or tools related to material formula or process performance are usually based on data-driven prediction models, such as neural networks, tree models, kernel methods, empirical equations, etc., which are used to output the prediction results of key performance such as material strength, hardness, elongation, etc., to assist engineers in judging the feasibility of formula or process and providing preliminary reference for subsequent experimental direction.

[0003] The existing material performance prediction related tools have obvious defects in actual application: only point prediction results of performance can be output, there is lack of stable confidence interval, engineers are difficult to evaluate the actual success grasp of performance meeting the standard; when facing the joint demand of multiple indexes, the overall success probability cannot be quantified; if the candidate formula or process exceeds the historical experience distribution, the prediction reliability will be greatly reduced, which is easy to mislead the experimental direction; and the reality of limited experimental resources is not considered, the experimental priority cannot be sorted according to success grasp, cost, etc., and the explanatory power is insufficient, engineers are difficult to locate the source of uncertainty and optimization direction, resulting in high trial and error cost and long research and development cycle, therefore, the present application provides a system and method for verifying material formula through confidence interval to solve such problems. SUMMARY

[0004] Technical problems to be solved

[0005] In view of the deficiencies of the prior art, the present application provides a system and method for verifying material formula through confidence interval to solve the problems proposed in the background art.

[0006] Technical scheme

[0007] To achieve the above purpose, the present application is implemented by the following technical scheme: a system for verifying material formula through confidence interval, as shown in Figure 1 , comprising:

[0008] A data collection module is configured to collect candidate formula basic data, historical experiment basic data, environment correlation data, material implicit data, and equipment state data, wherein the candidate formula basic data includes ingredient proportion and process parameters; the ingredient proportion is calculated by high-precision electronic balance weighing, with a weighing precision of 0.001 g, and a calculation method of single ingredient mass divided by total formula mass; the process parameters include reaction temperature and holding time, the reaction temperature is collected by an intelligent temperature control system, with a precision of plus or minus 0.5 degrees Celsius, and the holding time is collected by a timer, with a precision of plus or minus 1 second; the historical experiment basic data includes performance measured value, test repetition number, single measurement error, process parameter fluctuation standard deviation, and similar formula conventional ingredient proportion interval; the performance measured value is measured according to the material type by selecting a corresponding standard instrument, the strength of a metal material is measured by a universal testing machine, with a range of 0 to 1000 MPa and a precision of plus or minus 1 MPa, and the glass transition temperature of a plastic material is measured by a differential scanning calorimeter, with a precision of plus or minus 1 degree Celsius;

[0009] The test repetition number is determined according to the performance stability, and is usually 3-5 times, the performance which is easily affected by the environment is repeated 5 times, and the performance with good stability is repeated 3 times, and is recorded as an integer;

[0010] The single measurement error is the difference between the single performance measured value and the average value of multiple measurements, and the unit is consistent with the performance index;

[0011] The process parameter fluctuation standard deviation is calculated based on the process parameter data of similar historical formulas, and the calculation process is as follows: first, the number of process parameter data groups participating in the statistics is determined, then the difference between each single process parameter value and the average value of all data is calculated, each difference is squared and summed, the sum is divided by (the number of data groups minus one), and finally the result is squared, and the unit is consistent with the process parameter;

[0012] The similar formula conventional ingredient proportion interval is calculated based on the ingredient proportion data of similar historical formulas, and the calculation process is as follows: first, the average value and the standard deviation of the ingredient proportion in the similar historical formulas are calculated, the lower limit of the interval is obtained by subtracting 1.96 times the standard deviation from the average value, and the upper limit of the interval is obtained by adding 1.96 times the standard deviation to the average value;

[0013] The environment correlation data includes high-low temperature cycle attenuation rate, salt spray retention rate, and temperature and humidity difference; the high-low temperature cycle attenuation rate is obtained by testing with a high-low temperature cycle test chamber, the initial performance of the material is measured, the performance is measured after 100 cycles from-40 degrees Celsius to 85 degrees Celsius, and the calculation is (the initial performance minus the performance after the cycle) divided by the initial performance and multiplied by 100%;

[0014] The salt spray retention rate is tested by a salt spray test chamber, and after being sprayed with 5% sodium chloride solution for 48 hours, the calculation is (the performance after the test divided by the initial performance) multiplied by 100%;

[0015] The temperature difference is calculated by the difference between the highest temperature and the lowest temperature, and the humidity difference is calculated by the difference between the highest humidity and the lowest humidity;

[0016] The material implicit data includes raw material purity, particle size distribution standard deviation, and glass transition temperature, wherein the raw material purity is determined by a high-performance liquid chromatograph;

[0017] The particle size distribution standard deviation is determined by a laser particle size analyzer; the glass transition temperature is tested by a differential scanning calorimeter under a nitrogen atmosphere at a temperature rising rate of 10℃ / min, and the midpoint temperature of the sudden change interval of the differential scanning calorimetric curve is taken, with an accuracy of plus or minus 1℃;

[0018] The equipment state data includes calibration overage, repeated measurement standard deviation, key component running time, and zero drift, wherein the calibration overage is determined by comparison with a standard sample, i.e. the difference between the value measured by the equipment and the true value of the standard sample, with the same unit as the measurement parameter; the repeated measurement standard deviation is obtained by repeatedly measuring the same standard sample 10 times, and the calculation process is as follows: first, calculate the difference between each single measurement value and the average value of 10 measurements, square each difference, sum them up, divide the sum by (10-1), and finally take the square root of the result;

[0019] The key component running time is obtained by the equipment control system statistics or by a timer to accumulate the running time each time; the zero drift is obtained by recording the initial zero value after the equipment is preheated for 30 minutes, and recording the zero value again after the equipment is continuously running for 4 hours, and calculating the difference between the two zero values;

[0020] The calibration overage is detected once a day to ensure that the equipment measurement accuracy is in real-time monitoring state; the repeated measurement standard deviation is detected once every 3 days to avoid short-term equipment fluctuation affecting error calculation; the key component running time is recorded in real time, and wear evaluation is performed once every 500 hours of cumulative time; the zero drift is detected once before each equipment start, and the initial zero value is recorded after preheating for 30 minutes to ensure that the equipment zero point is not biased before each experiment;

[0021] The collected data is preprocessed;

[0022] The specific steps of preprocessing the data are as follows:

[0023] First, the total number of samples and the average value of the single-class historical data are counted, the difference between the single data value and the average value is calculated and squared, and the sum is divided by the result after the total number of samples is reduced by one, and the standard deviation is obtained by taking the square root of the result. The outlier data exceeding the preset multiple of the standard deviation is removed; the historical minimum and maximum values of the single-class data are determined, and the standardized data ranging from 0 to 1 is obtained by taking 0 as the fixed mapping lower limit, subtracting the historical minimum value from the original data, and then dividing the result by the difference between the historical maximum value and the historical minimum value. The cleaned and standardized data is mapped to a quantifiable value; a unique code is assigned to each candidate formula, and the environmental, material, and equipment related data are associated with the code and stored in a relational database, and finally the structured data set is output in CSV format, which includes the formula code, ingredient proportion, process temperature, raw material purity, equipment calibration error value, and environmental temperature and humidity difference.

[0024] The error calculation module is used to normalize the preprocessed environmental associated data, material implicit data and equipment state data, and then perform error calculation and correction to obtain the final error value, and construct and output the adjusted final confidence interval according to the final error value;

[0025] The specific steps of error calculation and correction of the preprocessed environmental associated data, material implicit data and equipment state data are as follows:

[0026] First, the preprocessed structured data set is divided according to the preset proportion, the preset proportion is 6 training sets, 2 calibration sets and 2 validation sets, and the training set, calibration set and validation set are divided according to the ingredient proportion interval (such as 0-5%, 5-10%) and process temperature interval (such as 20-30℃, 30-40℃), to ensure that the sample distribution of the training set, calibration set and validation set is completely consistent with the original structured data set, and to avoid affecting the subsequent calculation due to data distribution deviation;

[0027] Then, the calibration set is clustered, the clustering algorithm is K-means, and the number of clusters K of the K-means algorithm is determined by the elbow rule: the within-cluster sum of squares (WCSS) is calculated when K is from 2 to 10, when K increases to 5, the WCSS curve appears an obvious inflection point, and the WCSS decreases by less than 10% when K continues to increase, so the optimal cluster number K is determined as 5; the clustering effect is verified by the silhouette coefficient, and the silhouette coefficient is 0.72 when K is 5, and the clustering dimensions are ingredient proportion and process temperature: the ingredient proportion is divided into intervals with an interval of 5%, such as 0-5% and 5-10%, and the process temperature is divided into intervals with an interval of 10℃, such as 20-30℃ and 30-40℃. The calibration set is divided into multiple sample groups with similar characteristics by the algorithm.

[0028] First, the performance prediction value and the performance measured value of each sample in each sample group are obtained, and the absolute value of the performance residual error is obtained by subtracting the performance measured value from the performance prediction value; then the distribution of all performance residual error absolute values in each sample group is counted, and the 95% percentile is taken, that is, 95% of the performance residual error absolute values in the sample are less than or equal to the value, to obtain the basic error value.

[0029] In combination with the implicit data of the material, the purity and particle size distribution standard deviation parameters of the raw material are introduced, and the implicit correction coefficient of the material is obtained by comprehensive calculation, that is, in combination with the empirical coefficient based on historical data statistics, for example, by analyzing the experimental data of similar materials in the past three years: for every 1% decrease in the purity of the raw material, the performance error increases by 0.01 times on average; for every 1 micron increase in the particle size distribution standard deviation, the performance error increases by 0.005 times on average, the purity of the raw material and the particle size distribution standard deviation are substituted into the calculation, and the implicit correction coefficient of the material is obtained.

[0030] According to the equipment state data, the equipment state is determined, and the corresponding correction is applied to the basic error according to the determination result; based on the environmental correlation data, the high-low temperature cycle attenuation rate, the salt spray retention rate and the temperature and humidity difference parameters are introduced, and the environmental correction coefficient is obtained by comprehensive calculation; specifically, the calibration error value is determined by comparing the standard sample, that is, the difference between the value measured by the equipment and the true value of the standard sample, and the unit is consistent with the measurement parameter; the running time of the key components is read in real time from the equipment control system.

[0031] If the calibration error value is less than or equal to 0.5 and the running time of the key components is less than or equal to 1000 hours, the equipment state is determined to be excellent; if the calibration error value is greater than 0.5 and less than or equal to 1.0, or the running time of the key components is greater than 1000 hours and less than or equal to 2000 hours, the equipment state is determined to be medium; if the calibration error value is greater than 1.0, or the running time of the key components is greater than 2000 hours, the equipment state is determined to be poor.

[0032] When the state is excellent, the correction coefficient is 1.0, when the state is medium, the correction coefficient is 1.2, and when the state is poor, the correction coefficient is 1.5, to obtain the equipment state correction coefficient.

[0033] The high-low temperature cycle attenuation rate is obtained by testing in a high-low temperature chamber, the test conditions are 40 degrees Celsius below zero to 85 degrees Celsius for 100 cycles, and the result is in percentage units; the salt spray retention rate is obtained by testing in a salt spray test chamber, the test conditions are 5% sodium chloride solution spraying for 48 hours, and the result is in percentage units; the temperature and humidity difference is obtained by continuously collecting data for 24 hours with a temperature and humidity sensor.

[0034] Combine the empirical coefficient based on the statistical experiment of environmental impact, which is obtained by analyzing the performance deviation of materials in different environments: for every 1% reduction in high and low temperature cycle attenuation rate, the performance error increases by an average of 0.008 times; for every 1% reduction in salt spray retention rate, the performance error increases by an average of 0.003 times; for every 1 degree Celsius increase in temperature and humidity difference, the performance error increases by an average of 0.002 times. Substitute the three environmental parameters into the calculation to obtain the environmental correction coefficient.

[0035] Fuse the basic error with various correction coefficients to obtain the final error value;

[0036] Use the verification set to check the coverage rate. If the preset target is not met, scale the final error value by a certain proportion until the coverage rate meets the requirements, and output the adjusted final confidence interval.

[0037] Multiply the basic error value by the material implicit correction coefficient, then by the equipment state correction coefficient, and finally by the environmental correlation correction coefficient to obtain the final error value.

[0038] Determine the interval range of the performance prediction value plus or minus the final error value, count the number of samples in the verification set whose performance measured value falls within this interval, divide this number by the total number of samples in the verification set to obtain the coverage rate, and the result is in percentage. The preset coverage rate target is 90%.

[0039] If the calculated coverage rate is less than 90%, calculate the scaling coefficient, which is equal to 90% divided by the actual coverage rate. Multiply the original final error value by this scaling coefficient to obtain the adjusted final error value. Again, use the verification set to check the adjusted coverage rate. If it still does not reach 90%, repeat the scaling step until the coverage rate meets the preset target of 90%. Take the performance prediction value plus or minus the adjusted final error value as the final confidence interval.

[0040] Interval adjustment module, used to calculate the similarity and material influence value of the candidate formula and historical samples, the specific steps of similarity calculation are as follows:

[0041] Traverse all components of the candidate formula, weigh each component by high-precision electronic balance, calculate the ratio of single component mass to total formula mass, and obtain the component proportion of the candidate formula. Collect the process temperature of the candidate formula through the intelligent temperature control system and the holding time through the timer to obtain the process parameters of the candidate formula. Extract historical sample data of similar materials from the historical experiment database to obtain the component proportion and process parameters of the historical samples, and normalize the component proportion and process parameters to ensure that they are in the same dimension. Based on fixed feature dimension, calculate the distance evaluation value between them, and combine the maximum historical distance to obtain the similarity through comprehensive calculation;

[0042] The specific steps of calculating the distance evaluation value are as follows:

[0043] First, calculate the difference between the candidate formula and the historical sample corresponding component proportion, square each difference and sum; then calculate the difference between the corresponding process parameters, square each difference and sum; add the two sum results, square the added result to get the distance evaluation value;

[0044] The distance evaluation value is obtained as follows:

[0045]

[0046] In the formula, The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows:

[0047] The specific steps of obtaining the similarity are as follows:

[0048] The distance evaluation value between the candidate formula and the historical sample is obtained, and the maximum value in the distance evaluation value is calculated as the maximum historical distance. The distance evaluation value between the candidate formula and the historical sample is calculated to obtain the similarity.

[0049] The similarity is obtained as follows:

[0050]

[0051] In the formula, The similarity is obtained as follows: The similarity is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows: The distance evaluation value is obtained as follows:

[0052] The specific steps of obtaining the material influence value are as follows:

[0053] Put the material into a high-low temperature chamber, set the cycle condition to be from minus 40 degrees Celsius to 85 degrees Celsius, and continuously cycle 100 times; test the key performance values of the material before and after the cycle, for example, strength, calculate the ratio of the performance value after the cycle to the performance value before the cycle, and then multiply by 100% to obtain the high-low temperature cycle attenuation rate. The thermal behavior curve of the material is tested by a differential scanning calorimeter, and the glass transition temperature is determined according to the inflection point of the curve. The target environment maximum temperature is determined according to the actual application scene of the material, for example, 85 degrees Celsius for automotive materials and 40 degrees Celsius for indoor decoration materials;

[0054] Calculate the difference between the glass transition temperature and the target environment maximum temperature, take the absolute value of the difference, and obtain the temperature difference;

[0055] Multiply the temperature difference by 0.5 to obtain the temperature difference influence value; subtract 100 from the high-low temperature cycle attenuation rate, multiply the result by 0.3 to obtain the attenuation rate influence value; subtract the sum of the temperature difference influence value and the attenuation rate influence value from 100 to obtain the material influence value. The 0.5 and 0.3 are empirical coefficients based on the statistics of material environmental adaptability experiments, which are dimensionless.

[0056] The specific steps of adjusting the interval according to the similarity comparison result and the material influence value are as follows:

[0057] Set two similarity threshold values, a first threshold value and a second threshold value, wherein the first threshold value corresponds to the 95th percentile of the similarity distribution of historical samples, and the second threshold value corresponds to the 99th percentile of the similarity distribution of historical samples, and compare the calculated similarity with the first threshold value and the second threshold value respectively; Specifically: extract the similarity data between all historical samples from the historical experiment database, and statistically analyze the distribution of these data; take the 95th percentile of the distribution as the first threshold value, and take the 99th percentile of the distribution as the second threshold value. Set the preset score to 60 points, the first ratio to 1.05, and the second ratio to 1.1; obtain the adjusted final error value constructed before.

[0058] When the similarity is greater than or equal to the first threshold value, it indicates that the candidate formula has a high degree of similarity with the historical samples, and the historical data has strong support for the formula, so there is no need to adjust the interval width, and the final confidence interval remains unchanged.

[0059] When the similarity is between the first threshold value and the second threshold value, it means that the similarity between the candidate formula and the historical samples is moderate, and further judgment needs to be made in combination with the material influence value:

[0060] If the material influence value is greater than or equal to the preset score at this time, it means that the comprehensive performance of the material is good, and the interval width is adjusted by the first ratio to appropriately expand the interval coverage.

[0061] If the material impact value is less than the preset score, it indicates that the comprehensive performance of the material is relatively weak, and the interval width is adjusted by a second proportion to further expand the interval coverage range to cope with potential risks.

[0062] When the similarity is less than the second threshold, it indicates that the candidate formula is less similar to the historical sample, and the existing historical data is insufficient to reliably evaluate the formula. A rejection conclusion is output, and a prompt is given to supplement the historical data of the characteristic region of the formula, such as the proportion interval of a specific component, experimental data under specific process parameter conditions, etc.

[0063] For the case of keeping unchanged, the initial confidence interval is directly taken as the target confidence interval.

[0064] For the case of adjusting the width, the adjusted interval width is calculated according to the corresponding proportion to obtain the target confidence interval.

[0065] For the rejection case, no confidence interval is output, only a rejection conclusion and a data supplement prompt are output.

[0066] Finally, the determined target confidence interval or related conclusion is output.

[0067] The stability evaluation module is configured to compare the candidate formula basic data with the historical experimental basic data in real time to obtain a stability determination result.

[0068] The specific steps for obtaining the stability determination result are as follows:

[0069] The proportion information of all components is extracted from the candidate formula basic data, including the proportion value of each component in the formula and the corresponding component type. The same type of formula data as the candidate formula is selected from the historical experimental basic database to obtain the proportion data of each corresponding component in these historical formulas, and the conventional component proportion interval of the same type of formula is determined through statistical analysis. At the same time, the key process parameters such as reaction temperature, holding time, and pressure are extracted from the candidate formula basic data; the parameter data of the same type of process is extracted from the historical experimental basic data for subsequent calculation of process parameter fluctuation standard deviation.

[0070] The proportion of each component in the candidate formula is compared with the conventional proportion interval of the corresponding component in the same type of formula in the historical experimental basic data one by one:

[0071] If the proportions of all components in the candidate formula fall within the corresponding conventional interval, it indicates that the component proportion of the formula conforms to the common range of historical formulas of the same type, and the component proportion is determined to be reasonable.

[0072] If the proportion of at least one component in the candidate formula exceeds the corresponding conventional interval, it indicates that the component proportion of the formula is significantly different from the historical formulas of the same type, and the component proportion is determined to be insufficient.

[0073] Based on the same type of process parameter data in the historical experimental basic data with the same dimension as the candidate formula process parameter, such as the process parameter of the candidate formula is the reaction temperature, only the reaction temperature data of all the same type of formula in the historical data is extracted, and the specific requirements are: the number of data groups participating in the statistics ≥ 300 groups, which meets the law of large numbers, ensuring that the standard deviation calculation result is reliable, and the historical process parameter fluctuation standard deviation under this dimension is obtained through statistical calculation.

[0074] Subsequently, the target process parameter value is extracted from the candidate formula basic data, the parameter value is subtracted from the average value of the above-mentioned historical same type of process parameter, and the absolute value of the calculated difference value is taken, which is the process parameter fluctuation value.

[0075] Finally, the process parameter fluctuation value is used as the dividend, the historical process parameter fluctuation standard deviation is used as the divisor, the division operation is performed, and then the operation result is multiplied by 100% to convert it into percentage form, and the final result is the impact rate of process parameter fluctuation on performance.

[0076] The low threshold is set to 50%, and if the impact rate of process parameter fluctuation on performance is less than or equal to 50%, it means that the deviation of the process parameter of the candidate formula from the historical same type of process parameter is small, and the impact on performance is within an acceptable range, and it is determined to be process stable.

[0077] If the impact rate of process parameter fluctuation on performance is greater than 50%, it means that the deviation of the process parameter of the candidate formula from the historical same type of process parameter is large, which may have a significant impact on performance, and it is determined to be process unstable.

[0078] Based on the component rationality determination result and the process stability determination result, comprehensive analysis is performed:

[0079] When the components are reasonable and the process is stable, it means that the component ratio of the formula and the process parameter are both in a reasonable range, and the overall stability is good, and it is determined to be stable and qualified.

[0080] When the components are reasonable but the process is unstable, or the component rationality is insufficient but the process is stable, it means that the formula has certain problems in the component ratio or the process parameter, but it is not out of control overall, and the overall stability is at a medium level, and it is determined to be stable and medium.

[0081] When the component rationality is insufficient and the process is unstable, it means that the component ratio and the process parameter of the formula both have obvious problems, and the overall stability is poor, and it is determined to be stable and unqualified.

[0082] Through the above comprehensive determination, the final stability determination result is obtained.

[0083] The stability determination result is taken as an auxiliary reference, combined with the similarity and preset threshold comparison result and the material influence value, and the initial confidence interval is adjusted according to the specific steps of subsequent interval adjustment, and finally the adjusted target confidence interval is output.

[0084] The credibility evaluation module is configured to obtain a basic standard evaluation value based on the target confidence interval through random sampling, specifically, the number of random sampling times is set to 300, obtain a feasibility evaluation value combined with the stability determination result, obtain a final credible evaluation value by fusing the basic standard evaluation value and the feasibility evaluation value, and adopt product fusion based on historical data verification, which can preferentially reflect the key influence of stability on experimental success, and output a decision suggestion.

[0085] The specific steps of obtaining the final credible evaluation value and outputting the decision suggestion are as follows:

[0086] Obtain the interval width related parameters, obtain the final confidence interval of the current candidate formula performance index from the error calculation module, calculate the final confidence interval width, extract the confidence interval width of the same type of formula from the historical experiment database, statistically obtain the historical maximum interval width, and then calculate the interval relative width term;

[0087] Obtain the similarity of the candidate formula and the historical sample from the interval adjustment module, obtain the final error value of the current candidate formula, extract the final error value of the same type of formula from the historical experiment database, statistically obtain the historical maximum error, and then calculate the measurement error proportion term, fuse the above parameters according to the weight to obtain the final credible evaluation value;

[0088] The final credible evaluation value is obtained in the following manner:

[0089]

[0090] In the formula, The final credible evaluation value is obtained in the following manner: The final credible evaluation value is obtained in the following manner: Map to a conservative score of 0-100, wherein 0.95 is a conservative scaling coefficient, which is verified based on historical data, so that the evaluation result is conservative and ensures the experimental success rate. When the initial evaluation value is 0.8, the conservative score is 76 after being multiplied by 0.95, which corresponds to an actual experimental success rate of ≥85%. When the initial evaluation value is 0.7, the conservative score is 66.5, which corresponds to an actual experimental success rate of ≥75%, ensuring that the evaluation result is stable and reliable. The value range is 0 to 1, and the larger the value, the better the candidate formula. The final credible evaluation value is obtained in the following manner: The final credible evaluation value is obtained in the following manner: The interval relative width term is calculated in the following manner: The similarity is obtained in the following manner: The final error value is obtained in the following manner: representing the historical maximum error, This step calculates the measurement error proportion item, 、 and respectively, the width weight factor, the similarity weight factor and the error weight factor, and the sum of the three weight factors is 1, for example, 0.4, 0.3, 0.3;

[0091] Three groups of processes were set up for testing, and the parameters selected for the candidate process were: reaction temperature 150℃, holding time 30 minutes, pressure 2.5MPa, heating rate 10℃ / min.

[0092] The parameters used in the comparative process 1 were: reaction temperature 180℃, holding time 20 minutes, pressure 3.0MPa, heating rate 15℃ / min.

[0093] The parameters used in the comparative process 2 were: reaction temperature 120℃, holding time 40 minutes, pressure 2.0MPa, heating rate 5℃ / min.

[0094] Group Interval relative width term Similarity Measurement error proportion term Final trust evaluation value Candidate process 0.75 0.85 0.8 0.785 Comparative process 1 0.5 0.6 0.5 0.525 Comparative process 2 0.6 0.7 0.65 0.63

[0095] According to the table data, the candidate process is better than the comparative process 1 and the comparative process 2 in the interval relative width item, the similarity, and the measurement error proportion item, and the final credible evaluation value reaches 0.785, which is much higher than 0.525 of the comparative process 1 and 0.63 of the comparative process 2, indicating that the candidate process has a better comprehensive performance in performance standard reliability, interval stability, and error controllability. Moreover, the candidate process corresponds to candidate A, the comparative process 1 corresponds to candidate B, and the comparative process 2 corresponds to candidate C. The evaluation threshold is set as follows: the high threshold is 80 and above, i.e., the final credible evaluation value is greater than or equal to 80; the medium threshold interval is 60 to 80 below, i.e., the final credible evaluation value is greater than or equal to 60 and less than 80; the low threshold interval is 40 to 60 below, i.e., the final credible evaluation value is greater than or equal to 40 and less than 60; and the low threshold is 40 below, i.e., the final credible evaluation value is less than 40. The above threshold is determined based on the corresponding relationship between the evaluation value and the actual experimental success rate in historical experimental data, for example, when the final credible evaluation value reaches 80, the actual experimental success rate of the historical similar formula reaches an average of 90% or above.

[0096] According to the size of the final credible evaluation value, the corresponding experimental priority suggestion is output:

[0097] If the final credible evaluation value reaches the high threshold, it means that the formula has high credibility and experimental value, and the priority experiment suggestion is output.

[0098] If the final trust evaluation value is in the medium threshold interval, it indicates that the formula has a certain degree of credibility and experimental value, but the priority is lower than the former, and the output is an alternative experiment suggestion.

[0099] If the final trust evaluation value is in the low threshold interval, it means that the credibility of the formula is low and the experimental risk is high, and the output is an exploratory experiment suggestion, suggesting that small-scale exploratory experiments be conducted.

[0100] If the final trust evaluation value is lower than the low threshold, it means that the credibility of the formula is extremely low and the possibility of experimental success is small, and the output is a temporary experiment suggestion, suggesting that the formula be optimized and improved first.

[0101] At the same time, combined with the calibration deviation value and zero drift in the equipment state data:

[0102] If the calibration deviation value exceeds the allowed range, output a suggestion to recalibrate the related equipment, which can be calibrated using standard samples.

[0103] If the zero drift is too large, output a suggestion to calibrate the zero point of the equipment.

[0104] In addition, according to the deviation of the candidate formula component proportion, process parameters and historical reasonable range:

[0105] If the component proportion exceeds the historical reasonable range, output a suggestion to adjust it to a reasonable interval.

[0106] If the process parameters deviate from the historical reasonable range, output a suggestion to optimize the process parameters to be within the reasonable range.

[0107] Online update module: When new experimental data is generated, compare its measured value with the prediction interval at the time of evaluation. If the measured value falls within the interval, supplement the data to the calibration set and validation set of the historical database, and incrementally update the estimate of the basic error value. If the measured value falls outside the interval, it is marked as an out-of-domain sample and supplemented to the specific out-of-domain sample set of the historical database for subsequent optimization of similarity threshold and interval expansion logic.

[0108] Monthly automatic calculation of validation set coverage (target ≥ 90%) and rejection rate (target ≤ 10%): If the coverage rate is lower than 88% for two consecutive months, adjust the scaling coefficient of the final error value to between 1.05-1.1, and after adjustment, verify it again with the validation set to ensure that the coverage rate returns to more than 90%. If the rejection rate is higher than 12% for two consecutive months, lower the second threshold, specifically from the 99th percentile to the 98th percentile, to reduce the misjudgment rate of out-of-domain samples.

[0109] After removing the interval adjustment module, the final confidence evaluation value of the candidate process fluctuates from ±5% to ±12%; after removing the stability evaluation module, the matching degree between the experimental recommendation and the actual success rate decreases from 85% to 60%, verifying the supporting role of each module on the reliability of evaluation.

[0110] A verification method for material formula by confidence interval, as shown in Figure 2 comprises the following steps:

[0111] Step one, collect candidate formula basic data, historical experiment basic data, environment correlation data, material implicit data and equipment state data, and pretreat the collected data;

[0112] Step two, error calculation and correction are performed on the pretreated environment correlation data, material implicit data and equipment state data, to obtain a final error value, and an adjusted final confidence interval is constructed and output according to the final error value;

[0113] Step three, the similarity of the candidate formula and the historical sample and the material influence value are calculated, the interval is adjusted according to the comparison result of the similarity and the preset threshold value and in combination with the material influence value, and a target confidence interval is output;

[0114] Step four, the candidate formula basic data and the historical experiment basic data are compared in real time to obtain a stability determination result;

[0115] Step five, a basic compliance evaluation value is obtained based on the target confidence interval through random sampling, a feasibility evaluation value is obtained in combination with the stability determination result, a final confidence evaluation value is obtained by fusing the basic compliance evaluation value and the feasibility evaluation value, and a decision recommendation is output.

[0116] Advantages

[0117] The present application has the following advantages:

[0118] (1) The verification system and method for material formula by confidence interval, by fusing the final error value and constructing the adjusted final confidence interval, solves the problem that the existing material performance prediction tool can only output point prediction results, lacks stable and reliable intervals, and engineers are difficult to evaluate performance compliance actual success grasp, and through verification set verification coverage and error scale scaling, ensures that the confidence interval meets the preset coverage requirement, makes the performance compliance success grasp quantifiable, and engineers no longer need to rely on experience judgment, improves the accuracy of performance evaluation.

[0119] (2) The material formula verification system and method through confidence interval solves the problem that existing tools cannot quantify the overall success probability when facing the joint demand of multiple indexes, and converts the multiple index joint evaluation result into an intuitive confidence score, so that engineers can quickly grasp the overall success of multiple indexes through the score without complex conversion, reducing the difficulty of multiple index evaluation.

[0120] (3) The material formula verification system and method through confidence interval solves the problem that existing tools greatly reduce the prediction reliability when the candidate exceeds the historical experience distribution, and easily mislead the experimental direction, and ensures the prediction reliability of the candidate within the historical experience range through differential interval adjustment, while the out-of-domain rejection mechanism can avoid the experimental direction deviation caused by unreliable prediction and reduce invalid experiments.

[0121] (4) The material formula verification system and method through confidence interval solves the problem that existing tools do not consider the limited reality of experimental resources and cannot sort the experimental priority according to the success grasp, and optimizes the feasibility evaluation value combined with the stability judgment result, so that the experiment sorting considers both success grasp and formula stability, engineers can allocate resources to candidates with high confidence and high stability, avoid wasting resources on low grasp experiments, and improve the utilization rate of experimental resources.

[0122] Of course, implementing any product of the present application does not necessarily require all the advantages described above. BRIEF DESCRIPTION OF DRAWINGS

[0123] Figure 1 The structure diagram of the material formula verification system through confidence interval of the present application;

[0124] Figure 2 The flowchart of the material formula verification method through confidence interval of the present application. DETAILED DESCRIPTION

[0125] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0126] Embodiment 1

[0127] To verify the practical application effect and operability of the material formula verification system and method by confidence interval, the present embodiment takes the development of rubber formula for automobile seal as the application scenario, and develops the whole process verification around the core requirements of rubber material performance standard reliability quantification and experimental priority decision.

[0128] In combination with the mechanical property requirements of rubber materials for automobile seals, and in reference to the industry standards such as GB / T528-2009 “Determination of Tensile Stress-Strain Properties of Vulcanized Rubber or Thermoplastic Rubber”, GB / T531.1-2008 “Vulcanized Rubber or Thermoplastic Rubber Indentation Hardness Test Method Part 1: Shore Hardness Tester Method”, the material performance target of the present embodiment is set as: strength not less than 20 MPa, hardness in the range of 65A plus or minus 2A, elongation not less than 420%.

[0129] To compare the influence of different process parameters and ingredient proportions on the rubber performance standard reliability, three groups of rubber formulas with different characteristics are selected for evaluation (including conventional process, high temperature short curing process, and low temperature long curing process), and the core process parameters and key ingredient proportions of each group are as follows:

[0130] The material performance target is set as: strength not less than 20 MPa, hardness in the range of 65A plus or minus 2A, elongation not less than 420%.

[0131] Three groups of rubber formulas are selected for evaluation, and the core process parameters of each group are as follows: the reaction temperature of candidate A is 150℃, the holding time is 30 minutes, the carbon black proportion is 32%, and the vulcanizing agent proportion is 1.2%; the reaction temperature of candidate B is 160℃, the holding time is 25 minutes, the carbon black proportion is 35%, and the vulcanizing agent proportion is 1.0%; the reaction temperature of candidate C is 180℃, the holding time is 20 minutes, the carbon black proportion is 40%, and the vulcanizing agent proportion is 0.8%.

[0132] Through the complete process of data collection, error calculation, interval adjustment, and reliability scoring, the evaluation conclusions of each group of formulas are obtained: the performance index confidence interval of candidate A is strength between 23MPa and 27MPa, hardness between 65A and 67A, and elongation between 425% and 440%, the reliability score is 76, and it is recommended to carry out experiments first; the performance index confidence interval of candidate B is strength 21MPa to 25MPa, hardness 64A to 66A, and elongation 410% to 425%, the reliability score is 65, and it is recommended to evaluate again after fine tuning; the performance index confidence interval of candidate C is strength 18MPa to 22MPa, hardness 60A to 64A, and elongation 390% to 410%, there is no reliability score, the conclusion needs to be rejected, and it is recommended to supplement sampling or perform simulation first.

[0133] In terms of explainability details, take candidate A as an example for uncertainty source decomposition: model error accounts for 30%, because the performance prediction model simplifies the complex kinetics of the rubber cross-linking reaction, resulting in a systematic deviation of 1.2 MPa between the predicted and measured values; measurement error accounts for 25%, the hardness gauge calibration exceeds the tolerance value by 0.3A, which is within the allowed range of 0.5A, but the repeatability standard deviation of elongation measurement is 5%, which accounts for 25% of the total uncertainty; out-of-domain expansion accounts for 20%, the similarity of candidate A to historical samples is 0.85, which is lower than the interval expansion starting point (95% quantile of historical similarity distribution), and the interval needs to be moderately widened to cover potential performance fluctuations; data sparsity accounts for 25%, the number of historical samples in the 1.2% sulfur agent interval is only 45, less than the recommended 50, resulting in limited accuracy of similarity calculation.

[0134] Extract the three samples with the highest similarity to candidate A from the historical database and view their actual performance: the first sample has a carbon black content of 31.8%, a sulfur agent content of 1.1%, a process parameter of reaction temperature 148°C and holding time 28 minutes, and actual performance of strength 25 MPa, hardness 66A, and elongation 430%, with all 10 experiments meeting the standard; the second sample has a carbon black content of 32.2%, a sulfur agent content of 1.3%, a process parameter of reaction temperature 152°C and holding time 32 minutes, and actual performance of strength 26 MPa, hardness 67A, and elongation 428%, with 7 out of 8 experiments meeting the standard; the third sample has a carbon black content of 31.5%, a sulfur agent content of 1.2%, a process parameter of reaction temperature 150°C and holding time 30 minutes, and actual performance of strength 24.5 MPa, hardness 65.5A, and elongation 425%, with all 12 experiments meeting the standard.

[0135] Targeted fine-tuning suggestions: candidate B's elongation interval lower limit of 410% is lower than the target value of 420%, it is recommended to extend the holding time from 25 minutes to 28 minutes, and reference the 32-minute holding time of the second adjacent sample, it is expected that the adjusted elongation interval can be improved to between 420% and 435%; candidate C's reaction temperature of 180°C exceeds the historical reasonable interval of 140°C to 170°C, and the carbon black content of 40% exceeds the historical reasonable interval of 28% to 35%, which is an out-of-domain sample, it is recommended to first conduct simulation experiments under the conditions of reaction temperature 170°C and carbon black content 35%, and then evaluate after supplementing the performance data in this region.

[0136] Experimental decision logic: the performance indicators of candidate A are mostly within the target range, the strength lower limit 23MPa is greater than 20MPa, the hardness 65A to 67A meets the requirement of 2A up and down, the elongation lower limit 425% is greater than 420%, and the three most similar historical samples are all successful, so the experiment is preferred to be carried out; candidate B only has the elongation interval lower limit 410% lower than the target value 420%, and can be quickly optimized by fine-tuning the holding time, so it is suggested to fine-tune and evaluate again; the strength, hardness and elongation interval lower limit of candidate C are all lower than the target requirement, the strength 18MPa is less than 20MPa, the hardness 60A is less than 63A, the elongation 390% is less than 420%, and the process parameters are significantly beyond the historical reasonable range, which is an out-of-domain sample, so the direct experiment is high-risk, and it is necessary to supplement data first.

[0137] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0138] The preferred embodiments of the application disclosed above are only used to help explain the application. The preferred embodiments do not describe all the details of the application, nor limit the application to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of the specification. The specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. A system for verifying material formulations using confidence intervals, characterized in that, include: The data acquisition module is used to collect basic data of candidate formulations, basic data of historical experiments, environmental correlation data, implicit data of materials and equipment status data, and to preprocess the collected data. The error calculation module is used to calculate and correct the errors in the preprocessed environmental correlation data, material implicit data and equipment status data to obtain the final error value. Based on the final error value, the module constructs and outputs the adjusted final confidence interval. The specific steps for error calculation and correction of the preprocessed environmental correlation data, material implicit data, and equipment status data are as follows: The preprocessed data is formed into a structured dataset and divided into a training set, a calibration set, and a validation set according to a preset ratio. Based on the clustering of the calibration set according to the component proportion range and the process temperature range, a specific quantile of the absolute value of the performance residual of each class is calculated as the basic error value; By combining implicit data of materials and introducing parameters such as raw material purity and particle size distribution standard deviation, the implicit correction coefficient of materials is obtained through comprehensive calculation. The equipment status is determined based on the equipment status data, and corresponding corrections are applied to the basic errors based on the determination results; Based on environmental correlation data, parameters such as high and low temperature cycle attenuation rate, salt spray retention rate and temperature and humidity difference are introduced, and environmental correction coefficients are obtained through comprehensive calculation. The final error value is obtained by combining the basic error with various correction coefficients. The coverage is verified using the validation set. If the preset target is not met, the final error value is scaled proportionally until the coverage meets the requirements. The adjusted final confidence interval is then output. The interval adjustment module is used to calculate the similarity and material influence value between the candidate formula and historical samples. Based on the comparison results of similarity and preset threshold and combined with the material influence value, the interval is adjusted and the target reliable interval is output. The stability assessment module is used to compare the basic data of candidate formulations with historical experimental data in real time to obtain the stability judgment results. The credibility assessment module is used to obtain a basic compliance assessment value based on the target credibility interval through random sampling, obtain a feasibility assessment value by combining the stability judgment result, and obtain a final credibility assessment value by merging the basic compliance assessment value and the feasibility assessment value, and output decision recommendations.

2. The system for verifying material formulations using confidence intervals according to claim 1, characterized in that: The basic data for the candidate formulations include component proportions and process parameters; Historical experimental data includes measured performance values, number of test repetitions, single measurement error, standard deviation of process parameter fluctuations, and the range of common ingredient proportions in similar formulations. Environmental data include high and low temperature cycling degradation rate, salt spray retention rate, and temperature and humidity difference; Implicit material data includes raw material purity, particle size distribution standard deviation, and glass transition temperature; Equipment status data includes calibration deviation, standard deviation of repeated measurements, running time of key components, and zero drift.

3. The system for verifying material formulations using confidence intervals according to claim 1, characterized in that: The specific steps for preprocessing the collected data are as follows: Acquire basic data of candidate formulations, basic data of historical experiments, environmental correlation data, implicit data of materials, and equipment status data. Clean the data to remove outliers and eliminate outliers that exceed a preset multiple of the standard deviation. Standardize the cleaned data into a unified format and map it into directly calculable quantitative values. Perform correlation matching on the standardized data to bind the correspondence between the formulation and the corresponding environmental, material, and equipment data to obtain a structured dataset.

4. The system for verifying material formulations using confidence intervals according to claim 1, characterized in that: The specific steps for calculating the similarity between candidate formulations and historical samples are as follows: Obtain the component proportions and process parameters of candidate formulations and historical samples, calculate the distance evaluation value between them based on fixed feature dimensions, and obtain the similarity by combining the maximum historical distance.

5. The system for verifying material formulations using confidence intervals according to claim 1, characterized in that: The specific steps for obtaining the material influence value are as follows: Obtain the high and low temperature cycling decay rate, glass transition temperature, and the highest temperature of the target environment. Calculate the temperature difference between the glass transition temperature and the highest temperature of the target environment. Use this temperature difference and the high and low temperature cycling decay rate to obtain the material influence value through comprehensive calculation.

6. The system for verifying material formulations using confidence intervals according to claim 1, characterized in that: The specific steps for adjusting the range based on the similarity comparison result with the preset threshold and the material influence value are as follows: Two similarity thresholds are set: a first threshold and a second threshold. The first threshold corresponds to the high percentile of the historical sample similarity distribution, and the second threshold corresponds to the even higher percentile of the historical sample similarity distribution. The calculated similarity is then compared with the first threshold and the second threshold respectively. If the similarity is greater than or equal to the first threshold, the final confidence interval remains unchanged. If the similarity is between the first threshold and the second threshold and the material influence value is greater than or equal to the preset score, the interval width is adjusted according to the first ratio. If the similarity is between the first threshold and the second threshold and the material influence value is less than the preset score, the interval width is adjusted according to the second ratio. If the similarity is less than the second threshold, output a rejection conclusion and prompt for supplementary historical data of the basic data of the formula. Finally, output the adjusted target confidence interval.

7. The system for verifying material formulations using confidence intervals according to claim 1, characterized in that: The specific steps for obtaining the stability determination result are as follows: Extract the component proportions from the candidate formulation's basic data and compare them with the common component proportion ranges of similar formulations in historical experimental basic data. If the component proportions are within the common range, the component proportions are deemed reasonable; otherwise, the component proportions are deemed insufficiently reasonable. Extract process parameters from candidate formulation basic data, compare them with the standard deviation of process parameter fluctuations in historical experimental basic data, calculate the impact rate of process parameter fluctuations on performance, and determine the process is stable if the impact rate is less than or equal to the lower threshold; otherwise, determine the process is unstable. If the proportions are reasonable and the process is stable, it is determined that the stability is qualified; If the component is reasonable but the process is unstable, or if the component is not reasonable but the process is stable, it is judged as having moderate stability. If the proportions are not reasonable and the process is unstable, the stability is deemed unqualified, and the overall stability judgment result is obtained.

8. The system for verifying material formulations using confidence intervals according to claim 1, characterized in that: The specific steps by which the credibility assessment module obtains the final credibility assessment value and outputs decision recommendations are as follows: Perform random sampling a preset number of times within the target confidence interval, and statistically analyze the percentage of times multiple indicators simultaneously hit the preset target to obtain the basic compliance assessment value; Based on the stability assessment results output by the stability assessment module, different levels of feasibility assessment values ​​are assigned to different assessment scenarios; if the stability assessment result is qualified, the highest level of feasibility assessment value is assigned. If the rating is medium, assign a medium-level feasibility assessment value. If the result is unqualified, the lowest level of feasibility assessment value will be assigned. The final credible assessment value is obtained by integrating the basic compliance assessment value and the feasibility assessment value. The experiment priority is output based on the final credibility assessment value: if the credibility score reaches the high threshold, the priority experiment suggestion is output; if it is in the medium threshold range, the alternative experiment suggestion is output; if it is in the low threshold range, the exploratory experiment suggestion is output. If the value is below the lower threshold, a suggestion to postpone the experiment will be output. At the same time, based on the calibration deviation and zero drift in the equipment status data, a device calibration suggestion will be output. Based on the proportion of candidate formulation components and the deviation of process parameters from the historical reasonable range, a parameter adjustment scheme will be output.

9. A method for verifying material formulations using confidence intervals, applied to the material formulation verification system using confidence intervals as described in any one of claims 1-8, characterized in that: Includes the following steps: Step 1: Collect basic data of candidate formulations, basic data of historical experiments, environmental correlation data, implicit data of materials, and equipment status data, and preprocess the collected data; Step 2: Perform error calculation and correction on the preprocessed environmental correlation data, material implicit data and equipment status data to obtain the final error value, and construct and output the adjusted final confidence interval based on the final error value; Step 3: Calculate the similarity and material influence value between the candidate formulation and historical samples. Based on the comparison results of the similarity and the preset threshold, and combined with the material influence value, adjust the range and output the target confidence range. Step 4: Compare the basic data of the candidate formulation with the basic data of historical experiments in real time to obtain the stability determination results; Step 5: Based on the target confidence interval, obtain the basic compliance assessment value through random sampling, combine it with the stability judgment result to obtain the feasibility assessment value, integrate the basic compliance assessment value and the feasibility assessment value to obtain the final confidence assessment value, and output decision recommendations.

Images