Heat treatment quality pre-judgment analysis system and pre-judgment analysis method

Abstract

The invention relates to a heat treatment quality pre-judgment analysis system and a heat treatment quality pre-judgment analysis method. The system comprises a data processing module, a heat treatment quality pre-judgment module, a pre-judgment result evaluation module, a matching recommendation module and an optimization recommendation module. The method applies the system, the information of the workpiece to be pre-judged is input, the data processing module processes the information of the workpiece to be pre-judged to form a data set, the heat treatment quality pre-judgment module outputs a quality pre-judgment result according to the data set, and the evaluation module evaluates the quality pre-judgment result. The matching recommendation module outputs the details of the products with the matching degrees with the product characteristic parameters of the workpiece to be pre-judged in the process sample library, and the optimization recommendation module outputs an optimally matched heat treatment process data model according to the emphasis requirements of the selected workpiece. According to the method, the historical information data of the heat treatment process is collected, large model learning is performed, the heat treatment result is pre-judged according to the provided workpiece information, and the problem that the product quality is difficult to accurately pre-judge in the heat treatment process is solved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a heat treatment quality prediction and analysis system and a prediction and analysis method. Background Art

[0002] As the manufacturing industry shifts from traditional experience-based to modern digital, model-based, and intelligent, it is an inevitable trend to use industrial databases, data analysis, and process simulation technologies to solve many quality problems in production at low cost and high quality. As a key link affecting product quality, the heat treatment process has a complex quantitative relationship between process data and product quality, which is difficult to determine efficiently and accurately by traditional experience-based methods.

[0003] Therefore, based on big data modeling, the present invention provides a heat treatment quality prediction and analysis system, which makes full use of the technology of cognitive intelligence in massive data processing, efficiently solves the pain point that it is difficult to accurately predict the quality of the product during the heat treatment process, and helps to control the quality of the heat treatment. Summary of the invention

[0004] In view of the above problems, the present invention proposes a heat treatment quality prediction analysis system and a prediction analysis method, and the specific technical scheme is as follows:

[0005] A heat treatment quality prediction and analysis system is constructed in the following steps:

[0006] S1. Collect historical data and build a thermal processing data model based on the corresponding relationship between product chemical composition, process parameters and mechanical properties;

[0007] S2. Build a data processing module based on the principal component analysis algorithm to determine the key quality influencing factors and extract the process parameter bandwidth, process the hot processing process data model and establish a process sample library;

[0008] S3. Construct a heat treatment quality prediction module based on the random forest algorithm. The specific construction steps are as follows:

[0009] S3.1, randomly select n samples from the process sample library using a sampling method with replacement as the training set for a single decision tree;

[0010] S3.2, build a decision tree; starting from the root node, split the training data set into two sub-data sets according to the selected optimal splitting feature and threshold; for each sub-data set, recursively perform the splitting process until the maximum depth of the tree is 5;

[0011] S3.3, repeat the process of training set selection and decision tree construction to obtain multiple decision trees, which together constitute a random forest;

[0012] S3.4. Input a new data set, take the average of multiple decision tree prediction values, and output the heat treatment quality prediction result;

[0013] S3.5, repeat S2.1-S2.4 until the prediction result meets the requirements and the model weight is saved;

[0014] S4. Construct a quality prediction result evaluation module based on the comprehensive evaluation method, take the performance parameters of the quality prediction results as evaluation indicators, normalize the original quality prediction result data set, convert the indicator values ​​of different dimensions into dimensionless scores, and obtain the comprehensive evaluation score of the heat treatment quality prediction results;

[0015] S5. Construct a matching recommendation module based on the KNN nearest neighbor algorithm, and output the top five product details in the process sample library that are consistent with the feature parameters of the workpiece to be predicted;

[0016] S6. Build a recommendation module, input the workpiece’s performance focus, match it in the process sample library, and output the hot working process data model with the highest matching degree.

[0017] Further, S2 includes the following steps:

[0018] S2.1. Standardize the characteristics of heat treatment process parameters and convert the data into a distribution with a mean of 0 and a standard deviation of 1;

[0019] S2.2, for n heat treatment process parameter characteristics, construct an n×n covariance matrix;

[0020] S2.3, perform eigendecomposition on the covariance matrix C to obtain eigenvectors and eigenvalues;

[0021] S2.4, sort according to the size of the eigenvalues, and select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as the principal components;

[0022] S2.5, using the selected 10 principal components as new coordinate axes, projecting the original heat treatment process parameter data onto these coordinate axes to obtain new data sets and key influencing factors;

[0023] S2.6. Perform statistical analysis on the maximum, minimum, mean, and standard deviation of the data after dimensionality reduction to determine the data range bandwidth of key process parameters.

[0024] Furthermore, in S3.2, when each node of the decision tree is split, 10 features are randomly selected from all the features, and then an optimal split feature is selected from these 10 features for splitting.

[0025] Furthermore, in S4, for physical workpieces that have already started production, the deviation of the execution process data from the preset process data is used as a deduction item; for physical workpieces that have not started production and virtual workpieces, the degree of matching between feature information and data in the process data model is also used as a scoring item.

[0026] Furthermore, in S5 , the Euclidean distance between the process parameters of the input workpiece and each sample in the process sample library is calculated.

[0027] Furthermore, in S6, the user selectively inputs workpiece information and performance requirements, and the system matches the input conditions with the information in the process sample library, and outputs the best matching thermal processing process data model based on the weight value and matching degree.

[0028] A heat treatment quality prediction and analysis method uses the above-mentioned heat treatment quality prediction and analysis system, inputs the information of the workpiece to be predicted, the data processing module processes the information of the workpiece to be predicted to form a data set, the heat treatment quality prediction module outputs the quality prediction result according to the data set, the evaluation module evaluates the quality prediction result, the matching recommendation module outputs the details of the top five products in the process sample library that are ranked in the degree of matching with the product characteristic parameters of the workpiece to be predicted, and the optimization suggestion module outputs the best matching heat treatment process data model according to the emphasis requirements of the selected workpiece.

[0029] The beneficial effects of the present invention are:

[0030] The present invention collects historical information data of heat treatment process, performs large model learning, and predicts the heat treatment results according to the provided workpiece information, thereby solving the problem that it is difficult to accurately predict the product quality during the heat treatment process and facilitating the control of heat treatment quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] 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 or the description of the prior art 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.

[0032] Figure 1 is a schematic diagram of the heat treatment quality prediction analysis method of the present invention;

[0033] Figure 2 It is a schematic diagram of the actual workpiece processing of the present invention;

[0034] Figure 3 It is a schematic diagram of the processing of virtual workpieces according to the present invention;. DETAILED DESCRIPTION

[0035] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0036] The present invention provides the following specific embodiments:

[0037] In a first aspect, the present invention provides a heat treatment quality prediction and analysis system, the construction steps of which are as follows:

[0038] S1. Data collection: Collect a large amount of multi-dimensional data from actual production, including production process data, process data, quality data, equipment data, etc., to fully reflect various factors of the heat treatment process.

[0039] The collected multi-dimensional historical data is deposited in the data center. In accordance with the data governance concepts and requirements of DAMA and DCMM, guided by data planning and management consulting methods, the data architecture design and data model design of hot working process parameters are carried out to achieve unified collection and management of multi-source heterogeneous data and complete the construction of the hot working process parameter data warehouse. According to data standards, data quality, data security, metadata management and other governance activities, the hot working process parameter data assets are deposited, data subscription and service release based on asset catalogs are supported, and systematic process monitoring is realized. It also has diversified analysis capabilities for business, data and technology, and can build enterprise-level service capabilities, and further build a large database based on platform data.

[0040] Based on the data middle platform and the correspondence between the chemical composition, process parameters and mechanical properties of historical products, a heat treatment process data model is established, and a related type process sample library is established. This sample library contains the heat treatment process parameters and corresponding mechanical properties data of different types of workpieces, providing a basis for subsequent data analysis and quality prediction.

[0041] S2. Build a data processing module based on the principal component analysis algorithm to determine the key quality influencing factors and extract the process parameter bandwidth, process the hot working process data model and establish a process sample library.

[0042] The specific steps are as follows: S2.1. Standardize the characteristics of the heat treatment process parameters, that is, convert the data into a distribution with a mean of 0 and a standard deviation of 1. The calculation formula is as follows:

[0043]

[0044] Where μ is the mean of all sample data, and σ is the standard deviation of all sample data.

[0045] S2.2. For n heat treatment process parameter characteristics, construct an n×n covariance matrix C, where each element C ij Represents the covariance between feature i and feature j, assuming that the heat treatment process parameter set is a three-dimensional data set of x, y, and z. The covariance matrix is ​​a 3*3 matrix as shown in the following figure:

[0046]

[0047] If the covariance is positive, the two parameters of the heat treatment process are positively correlated. If the covariance is negative, the two parameters of the heat treatment process are negatively correlated.

[0048] S2.3. Perform eigendecomposition on the covariance matrix C to obtain eigenvectors and eigenvalues.

[0049] S2.4. Sort by eigenvalue and select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as principal components.

[0050] S2.5. Use the selected 10 principal components (i.e., eigenvectors) as new coordinate axes, project the original heat treatment process parameter data onto these coordinate axes, and obtain new data sets and key influencing factors.

[0051] S2.6. Perform statistical analysis on the maximum, minimum, mean, and standard deviation of the data after dimensionality reduction to determine the data range bandwidth of key process parameters.

[0052] S3. Construct a heat treatment quality prediction module based on the random forest algorithm. The specific construction steps are as follows:

[0053] S3.1. Use sampling with replacement to randomly select n samples from the process sample library as the training set for a single decision tree.

[0054] S3.2. Construct a decision tree; starting from the root node, split the training data set into two sub-data sets according to the selected optimal splitting feature and threshold; for each sub-data set, recursively perform the splitting process until the maximum depth of the tree is 5. When each node of the decision tree is split, 10 features are randomly selected from all features, and then the best splitting feature is selected from these 10 features for splitting.

[0055] S3.3. Repeat the process of training set selection and decision tree construction to obtain multiple decision trees, which together constitute a random forest.

[0056] S3.4. Input a new data set, take the average of the prediction values ​​of multiple decision trees, and output the heat treatment quality prediction result.

[0057] S3.5. Repeat S2.1-S2.4 until the prediction result meets the requirements and the model weight is saved.

[0058] S4. Construct a quality prediction result evaluation module based on the comprehensive evaluation method. Take the performance parameters of the quality prediction results such as hardness, impact toughness, fatigue strength, metallographic structure, decarburization depth, deformation, cracking, overheating and overburning as evaluation indicators. Normalize the original quality prediction result data set, convert the indicator values ​​of different dimensions into dimensionless scores, and obtain the comprehensive evaluation score of the heat treatment quality prediction result. For physical workpieces that have started production, the deviation of the execution process data from the preset process data is used as a deduction item; for physical workpieces that have not started production and virtual workpieces, the degree of matching between the feature information and the data in the process data model is also used as a scoring item. Add the scores of all indicators to obtain the comprehensive evaluation score of the heat treatment quality prediction result.

[0059] The final output is the quality prediction results, including performance prediction results, actual quality prediction scores and effect evaluation, as well as items with the most lost points.

[0060] S5. Construct a matching recommendation module based on the KNN nearest neighbor algorithm, and output the top five product details in the process sample library that match the characteristic parameters of the workpiece to be predicted. Specifically, for the process parameter input sample, including material grade, chemical composition, process information (such as quenching / normalizing temperature, holding time, cooling method, sample type), calculate the Euclidean distance between the input process parameter and each sample in the historical thermal process parameter data set. The formula is as follows:

[0061]

[0062] Assume K=5, that is, output the details of the five products with the highest degree of consistency in product characteristic parameters. The system supports direct viewing of their main characteristic parameters, and after clicking, you can view the actual process and quality information of the entire process.

[0063] S6. Construct a recommendation module, input the workpiece's performance focus, match it in the process sample library, and output the hot working process data model with the highest matching degree. Specifically, the user can choose to input one of the material grade or chemical composition, and one of the hardness requirements or performance requirements (including sample type), or enter all the condition items. The system matches the input conditions with the information in the process sample library, and outputs the best matching hot working process data model based on the weight value and matching degree.

[0064] Second, as Figure 1 As shown, the present invention proposes a heat treatment quality prediction and analysis method, and applies the above-mentioned heat treatment quality prediction and analysis system, and inputs the information of the workpiece to be predicted, the workpiece information includes material brand, chemical composition, process information (quenching / normalizing temperature, holding time, cooling method, sample type), part number (only for physical workpieces), hardness and physical and chemical performance standards and other information. The data processing module processes the workpiece information to be predicted to form a data set, the heat treatment quality prediction module outputs the quality prediction result according to the data set, the evaluation module evaluates the quality prediction result, the matching recommendation module outputs the details of the top five products in the process sample library that are consistent with the product characteristic parameters of the workpiece to be predicted, and the recommendation module outputs the best matching heat treatment process data model according to the focus requirements of the selected workpiece.

[0065] like Figure 2 As shown in the figure, the input workpiece is the actual workpiece (the workpiece that has started production and the workpiece that has not started production), and the process production information of the physical workpiece that has been carried out is obtained; the nodes that have been executed are displayed when the prediction results are output; the input information and actual execution information of the physical workpiece are systematically matched with the actual typical product data bandwidth information in the process data model; the quality prediction result information such as hardness results and mechanical properties results are output; the prediction results are comprehensively scored; the details of the five products with the highest degree of consistency in product characteristic parameters are output; click to view the actual process and quality information of the whole process; the predicted hardness and physical and chemical properties results are output. The results are compared with the standard requirements to determine whether the requirements are met; the deviation of the execution process data from the preset process data is used as a deduction item; the final quality prediction result is output; the items that lose the most points are output; the optimization suggestions are output: input the brand, chemical composition, hardness requirements, and performance requirements of the physical workpiece; match the input conditions with the typical category material / hardness / performance data bandwidth information in the process optimization model; the best matching typical category furnace matching process and process execution information can be used as the basic data for process optimization; output optimization process parameters, such as quenching / normalizing temperature range, holding time, cooling method, tempering temperature range, holding time, etc.

[0066] like Figure 3As shown in the figure, the input workpiece is a virtual workpiece (a virtual workpiece of different types that has not been manufactured but has quality prediction requirements), and the material brand, chemical composition, virtual process information (quenching / normalizing temperature, holding time, cooling method, sample type), hardness, and physical and chemical performance standards of the virtual workpiece are input; the input information is systematically matched with the actual typical product data bandwidth information in the process data model or the calculated data bandwidth information; the hardness result, mechanical property result and other quality prediction result information are output; the prediction results are comprehensively scored; the details of the five products with the highest degree of consistency in product characteristic parameters are output; after clicking, the actual process and quality information of the whole process can be viewed; the prediction The determined hardness and physical and chemical properties are compared with the standard requirements to determine whether the requirements are met; the degree of matching between the feature information and the data in the process data model is also used as a scoring item; the final quality prediction result is output; the items with the most lost points are output; and optimization suggestions are output: input the material brand, chemical composition, hardness requirements, and performance requirements of the virtual workpiece; match the input conditions with the typical category material / hardness / performance data bandwidth information in the process optimization model; the best matching typical category furnace matching process and process execution information can be used as the basic data for process optimization; output optimization process parameters, such as quenching / normalizing temperature range, holding time, cooling method, tempering temperature range, holding time, etc.

[0067] By collecting historical information data of the heat treatment process, conducting large-scale model learning, and predicting the heat treatment results based on the provided workpiece information, the problem of it being difficult to accurately predict product quality during the heat treatment process is solved, which helps control the quality of the heat treatment.

[0068] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A heat treatment quality prediction and analysis system, characterized in that: The construction steps are as follows, S1. Collect historical data and build a thermal processing data model based on the corresponding relationship between product chemical composition, process parameters and mechanical properties; S2. Build a data processing module based on the principal component analysis algorithm to determine the key quality influencing factors and extract the process parameter bandwidth, process the hot processing process data model and establish a process sample library; S3. Construct a heat treatment quality prediction module based on the random forest algorithm. The specific construction steps are as follows: S3.1, randomly select n samples from the process sample library using a sampling method with replacement as the training set for a single decision tree; S3.2, build a decision tree; starting from the root node, split the training data set into two sub-data sets according to the selected optimal splitting feature and threshold; for each sub-data set, recursively perform the splitting process until the maximum depth of the tree is 5; S3.3, repeat the process of training set selection and decision tree construction to obtain multiple decision trees, which together constitute a random forest; S3.

4. Input a new data set, take the average of multiple decision tree prediction values, and output the heat treatment quality prediction result; S3.5, repeat S2.1-S2.4 until the prediction result meets the requirements and the model weight is saved; S4. Construct a quality prediction result evaluation module based on the comprehensive evaluation method, take the performance parameters of the quality prediction results as evaluation indicators, normalize the original quality prediction result data set, convert the indicator values ​​of different dimensions into dimensionless scores, and obtain the comprehensive evaluation score of the heat treatment quality prediction results; S5. Construct a matching recommendation module based on the KNN nearest neighbor algorithm, and output the top five product details in the process sample library that are consistent with the feature parameters of the workpiece to be predicted; S6. Build a recommendation module, input the workpiece’s performance focus, match it in the process sample library, and output the hot working process data model with the highest matching degree.

2. A heat treatment quality prediction and analysis system according to claim 1, characterized in that: S2 includes the following steps, S2.

1. Standardize the characteristics of heat treatment process parameters and convert the data into a distribution with a mean of 0 and a standard deviation of 1; S2.2, for n heat treatment process parameter characteristics, construct an n×n covariance matrix; S2.3, perform eigendecomposition on the covariance matrix C to obtain eigenvectors and eigenvalues; S2.4, sort according to the size of the eigenvalues, and select the eigenvectors corresponding to the top 10 largest eigenvalues ​​as the principal components; S2.5, using the selected 10 principal components as new coordinate axes, projecting the original heat treatment process parameter data onto these coordinate axes to obtain new data sets and key influencing factors; S2.

6. Perform statistical analysis on the maximum, minimum, mean, and standard deviation of the data after dimensionality reduction to determine the data range bandwidth of key process parameters.

3. A heat treatment quality prediction and analysis system according to claim 1, characterized in that: In S3.2, when each node of the decision tree is split, 10 features are randomly selected from all features, and then an optimal split feature is selected from these 10 features for splitting.

4. A heat treatment quality prediction and analysis system according to claim 1, characterized in that: In S4, for physical workpieces that have already started production, the deviation of the execution process data from the preset process data is used as a deduction item; for physical workpieces that have not started production and virtual workpieces, the degree of matching between the feature information and the data in the process data model is also used as a scoring item.

5. A heat treatment quality prediction and analysis system according to claim 1, characterized in that: In S5 , the Euclidean distance between the process parameters of the input workpiece and each sample in the process sample library is calculated.

6. A heat treatment quality prediction and analysis system according to claim 1, characterized in that: In S6, the user selectively inputs workpiece information and performance requirements, and the system matches the input conditions with the information in the process sample library, and outputs the best matching hot working process data model based on the weight value and matching degree.

7. A heat treatment quality prediction and analysis method, using the heat treatment quality prediction and analysis system according to any one of claims 1 to 6, characterized in that: The information of the workpiece to be predicted is input, and the data processing module processes the information of the workpiece to be predicted to form a data set. The heat treatment quality prediction module outputs the quality prediction result according to the data set. The evaluation module evaluates the quality prediction result. The matching recommendation module outputs the details of the top five products in the process sample library that are consistent with the product characteristic parameters of the workpiece to be predicted. The optimization suggestion module outputs the best matching heat treatment process data model according to the emphasis requirements of the selected workpiece.

Images