Data-driven foaming product foaming cavity root cause analysis method and system
By applying decision tree learning model and big data technology in the foaming industry, a mapping relationship model between foaming process characteristics and hollow states was established, which solved the problem that traditional experimental methods were difficult to analyze the root cause of quality, and efficient hollow state prediction and root cause analysis were achieved, which improved the feasibility and accuracy of process improvement.
Patent Information
- Application Number
- CN202411614686.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-13
- Publication Date
- 2025-03-07
AI Technical Summary
In the manufacturing of foaming industrial products, traditional randomized testing methods are difficult to effectively analyze the quality root causes in the foaming process, resulting in a large amount of manpower, material resources and financial resources consumption, and some experiments are not operational at the time and space level.
The decision tree learning model technology is used to use the panoramic big data of the foaming process of foaming products to perform data preprocessing and structured sample construction, and a decision tree model is established to reflect the mapping relationship between the foaming process characteristics and the foaming hollow state, and to conduct hollow state prediction and root cause analysis.
Efficient prediction and root cause analysis of foam cavity state are achieved, feasibility, effectiveness and accuracy of foaming process improvement are improved, and the consumption of manpower and material resources is reduced.
Smart Images
Figure CN119578971A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data-driven analysis technology, and in particular to a data-driven foam product foaming cavity root cause analysis method and system. Background Art
[0002] Modern foaming industrial product manufacturing is changing from the traditional large-scale, standardized production model to a small-batch, multi-category, and highly flexible production model. Under this production model, the production process is highly complex, with many process variables and strong coupling with quality response, showing strong nonlinear, dynamic and time-varying characteristics, which often make the foaming industrial product process difficult to detect process defects and accumulate quality problems at the product terminal. Therefore, in each link of the foaming process, the prediction of quality conditions and the analysis of causes are particularly important. However, in this case, using the traditional random test method to analyze the root causes of quality requires the construction of a large number of parallel tests with different characteristic factor level combinations, which leads to a large amount of manpower, material and financial resources. Some tests are even inoperable at the time and space level. Summary of the invention
[0003] In order to solve the above-mentioned problems in the prior art, the present invention considers utilizing the panoramic big data of the foaming process of foaming products and the decision tree learning model technology to provide a data-driven foaming product foaming void state prediction and root cause analysis method and system for the foaming process, which predicts the void state and root cause analysis of the foaming link in the foaming product production process.
[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0005] A data-driven foaming product foaming void root cause analysis method comprises the following steps:
[0006] Step 1, collecting raw data of foaming cavity infrared detection and each step of foaming process;
[0007] Step 2: Perform data preprocessing on the collected raw data, and form a structured foam product foaming big data sample {foaming product ID, process characteristics of each link, foaming cavity state label} according to the timestamp;
[0008] Step 3, a) dividing the foaming big data samples of structured foaming products into a training set and a test set, and randomly sampling part of the sample data in the training set as a validation set; b) using the remaining data in the training set to build a decision tree training model, using the Gini index as the analysis basis, iteratively selecting the best features and feature value division points until the stop condition is met, and obtaining a mapping relationship model for reflecting the process characteristics of each link and the foaming void state label, and selecting hyperparameters based on the validation set; evaluating the model performance of the training model on the test data set based on the model evaluation index, and obtaining an ideal void state prediction model; c) using the trained decision tree ideal model to predict the foaming product with unknown void state;
[0009] Step 4: According to the mapping relationship between process characteristics and void status in the established model and the calculation result of the Gini index, the branching process of the decision tree subtree is determined to obtain the importance ranking of process characteristic factors affecting the void status and the characteristic rules that lead to the discrimination of different void status categories.
[0010] The collected data on each link of the foaming process of the foaming products include: the premixing stage, preheating stage, injection stage and foaming stage of the foaming process as well as the environment in which the process is located; the process information characteristics of each link include: foaming gun tip flow, gun tip pressure, foaming process temperature, material tank temperature, foaming injection amount, foaming agent ingredient ratio, preheating temperature, and environmental temperature, humidity and atmospheric pressure.
[0011] The data preprocessing operations include identification, cleaning, deduplication, and standardization.
[0012] In the data integration process, the void state is first determined based on the infrared detection results of the foaming voids, and the labeling is completed. The labeling result is whether the foaming void state of the current foaming product is qualified or unqualified; then, based on the foaming process timestamp and the foaming product ID as the index, a structured foaming product foaming big data sample {foaming product ID, process characteristics of each link, foaming void state label} is established; the foaming void state is used as the subsequent model output response, and other process characteristics are used as model feature inputs.
[0013] The sample size ratio of the training set and the test set fluctuates between 9:1 and 7:3, and the sample size ratio of the validation set in the training set to the training set fluctuates between 1:9 and 1:8.
[0014] Model hyperparameters include: the maximum number of branches of decision tree leaf nodes, the maximum depth of the tree, the minimum number of samples required for node partitioning, and the maximum number of leaf nodes; hyperparameter search methods include: grid search, random search, or Bayesian search; stopping conditions include: all samples contained in the current node belong to the same category, or the sample set contained in the current node is empty, or the current attribute set is empty, or all samples have the same values on all attributes and cannot be divided; model evaluation indicators include: error rate, accuracy, precision rate, recall rate, F1 value, AUC value, and ROC value, which are used to prune the model according to the evaluation indicator results to prevent overfitting.
[0015] The Gini index calculation formula is:
[0016] For a given sample set D, its Gini index is defined as:
[0017]
[0018] Where Gini represents the Gini index, D represents the current data set, and C k is the subset of samples in D that belong to the kth class;
[0019] If the sample set D is divided into two sample sets D1 and D2 by a certain feature a taking a certain value, then under the condition of the value of feature a, the Gini index calculation formula of set D is:
[0020]
[0021] Wherein, a represents a certain feature of the foaming process characteristics;
[0022] In the candidate attribute set A, select the attribute that minimizes the Gini index after partitioning:
[0023] a * =argmin[Gini(D,a)], where a∈A
[0024] Among them, a * It is used as the optimal partitioning attribute and clarifies the feature value partitioning points. It is also used as the basis for the subsequent quantification and ranking of the importance of feature contribution levels for classification problems. In addition, the generated decision tree branch structure is used to implement feature rules for distinguishing different void state categories.
[0025] By performing weighted cumulative calculation on the Gini coefficient values of different process characteristics at each branch node of the decision tree, the quantitative value of the importance level of the process characteristics' contribution to the foaming cavity state classification problem is obtained, and the importance of the process characteristics of each link is further ranked from large to small.
[0026] A data-driven foam product foaming void root cause analysis system, comprising a data acquisition module, an integrated processing module, a model building module and a root cause analysis module;
[0027] Data acquisition module, which collects the original data of foaming cavity infrared detection and each step of foaming process;
[0028] The integrated processing module performs data preprocessing operations on the collected raw data to construct a structured foam product foaming big data sample {foaming product ID, process characteristics of each link, foaming cavity state label};
[0029] Model building module, a) divide the big data samples of structured foaming products into training set and test set, and randomly sample some sample data in the training set as the validation set; b) use the remaining data in the training set to build a decision tree training model, and use the Gini index to iteratively select the best features and division points until the stop condition is met, and obtain a mapping relationship model that reflects the process characteristics of each link and the foaming cavity state label; and select hyperparameters based on the validation set; evaluate the training model based on the model evaluation index, and obtain the ideal model based on the model performance on the test data set; c) use the trained decision tree ideal model to predict the cavity state of unknown foaming products;
[0030] The root cause analysis module determines the branching process of the decision tree subtree based on the established model mapping relationship and the Gini index calculation results, and obtains the importance ranking of the process characteristic factors that affect the void state and the characteristic rules that lead to the discrimination of different void state categories.
[0031] The present invention has the following advantages and benefits:
[0032] The interpretability of the decision tree model is used in the root cause analysis of the foaming void state. It has the advantages of clear logic, high readability, convenience and efficiency, and easy iterative optimization. It can be used to analyze the causes of abnormal foaming states in the foaming process and to assist in the decision-making process of adjusting and optimizing working conditions, thereby improving the feasibility, effectiveness and accuracy of foaming process improvements.
[0033] At the same time, the analysis system provided by the present invention is a modular system with high compatibility and integration and a simple architecture. It can combine other different types of advantageous methods or algorithm interfaces of other systems in the data acquisition and integration links / modules to achieve efficient data processing. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 is a flow chart of the method of the present invention;
[0035] Figure 2 It is a recursive tree structure diagram obtained according to the method of the present invention. DETAILED DESCRIPTION
[0036] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the invention, so the present invention is not limited by the specific implementation disclosed below.
[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.
[0038] See also Figure 1 As shown, the present invention is a method and system for predicting the foaming cavity state and root cause analysis of foaming products based on a data-driven model. A specific application example includes the following contents:
[0039] A specific implementation system includes a data acquisition module, an integrated processing module, a model building module and a root cause analysis module. The data acquisition module is used to collect the infrared detection results of foaming voids and the original data of each link of the foaming process; the integrated processing module is used for data preprocessing of the original data and building an effective big data sample; the model building module is used to establish a decision tree model with the void state as the response and the process characteristics as the input, which is used to reflect the mapping mechanism between the model input and the response and realize the prediction of the void state; the root cause analysis module is used to provide the importance ranking of the process characteristic factors and the characteristic rules that lead to the discrimination of different void state categories. The specific implementation steps of each module jointly realizing a method for predicting the foaming void state and root cause analysis of foam products based on a data-driven model are as follows:
[0040] (1) Step S1: In the data acquisition module of the system, the original data of the infrared detection of foaming voids and each link of the foaming process are collected. The collected links of the foaming process of the foaming product include: the premixing stage, preheating stage, injection stage and foaming stage of the foaming process and the process environment, etc. The specific process information characteristics may include: the preheating temperature of the foaming product before foaming, the proportion of premixed catalyst, the actual content of the foaming agent, the flow rate of the foaming machine liquid material, the filling amount of the filler, the pressure of the filler metering pump, the ratio of isocyanate-combined polyether, the foaming pressure of the injection gun head, the injection temperature, flow rate and flow velocity of the isocyanate and combined polyether at the gun head, the temperature of the isocyanate and combined polyether working tank, the temperature and duration of the foaming process, and the atmospheric pressure, temperature and humidity of the environment where the foaming process equipment is located. The collection method may include the SCADA system of each process and detection link, the sensor or the process equipment PLC.
[0041] (2) Step S2: In the integrated processing module of the system, pre-processing operations such as data identification, cleaning, deduplication, and standardization are performed on the collected raw data. At the same time, the void state is identified based on the infrared detection result of the foam voids of the foam product, and the labeling is completed. The labeling result is the foam void state of the current foam product (whether the foaming process of the foam product with the current ID is qualified). Subsequently, based on the process timestamp and the foam product ID as the reference, the foam void state and the characteristic information of each process link are integrated, wherein the void state is used as the subsequent model output response, and other process characteristics are used as model input, so as to form a structured foam product foaming big data sample {foam product ID, process characteristics of each link, foam void state label}.
[0042] (3) Step S3: In the model building module of the system, the integrated foaming product big data samples are divided into a training set and a test set. At the same time, a portion of the sample data in the training set is randomly sampled as a validation set, wherein the sample size ratio of the training set to the test set is 8:2, and the sample size ratio of the validation set to the training set in the training set is 1:9.
[0043] (4) Step S4: In the model building module of the system, the remaining data in the training set is used to build a decision tree algorithm model, and the Gini index is used to select the best features and partitioning points to meet the stopping conditions, so as to obtain a mapping mechanism model between the model input and the model response, and select hyperparameters based on the validation set, wherein the decision tree model hyperparameters include: the maximum number of branches of the decision tree leaf nodes, the maximum depth of the tree, the minimum number of samples required for node partitioning, the maximum number of leaf nodes, etc.; the optional hyperparameter search methods include: grid search, random search or Bayesian search, etc.; the stopping conditions include: the samples contained in the current node all belong to the same category, or the sample set contained in the current node is empty, or the current attribute set is empty, or all samples have the same values on all attributes and cannot be divided; the Gini index calculation formula used is:
[0044] For a given sample set D, its Gini index is defined as:
[0045]
[0046] Where Gini represents the Gini index, D represents the current data set, and C k is the subset of samples in D that belong to the kth class.
[0047] If the sample set D is divided into two sample sets D1 and D2 based on whether a certain feature a takes a certain value, then under the condition of feature a, the Gini index calculation formula of set D is:
[0048]
[0049] In the candidate attribute set A, the attribute that minimizes the Gini index after partitioning is selected as the optimal partitioning attribute, that is, a * = argmin[Gini(D,a)], where a∈A. This can achieve the quantification of feature contribution levels for classification problems and subsequent sorting.
[0050] (5) Step S5: In the model building module of the system, the model performance of the training model on the test data set is evaluated based on the model evaluation indicators. The available evaluation indicators include: error rate, accuracy, precision rate, recall rate, F1 value, AUC value and ROC value, etc. According to the evaluation index results, the model is pruned to prevent overfitting; finally, the trained decision tree model is used to predict the foaming cavity state of the new foaming product process feature input, wherein the result of the foaming cavity state prediction is whether the cavity state is qualified, that is, the foaming cavity state is classified into qualified and unqualified.
[0051] (6) Step S6: In the root cause analysis module of the system, the branching process of the decision tree subtree is determined according to the established model mapping mechanism and the Gini index calculation result. Specifically, Figure 2 As shown in , taking the tree structure model containing three foaming process parameters as an example, firstly, whether the top plate temperature of the foaming process equipment is greater than 35.67°C is used as the threshold to divide it downward into a binary tree. On the right subtree of the binary tree, the sample void state has been classified. On the left subtree of the binary tree, whether the isocyanate gun head pressure of the foaming machine is greater than 132.05 bar is used as the threshold to divide it downward into the third layer of binary trees... until the termination condition is reached, the recursive tree structure generation is completed, and the importance ranking of process feature factors is provided (the Gini coefficient values of different process features at each branch node of the decision tree are cumulatively weighted to obtain the quantitative value of the process feature importance level, thereby realizing the importance sorting from large to small) and the feature rules that lead to the discrimination of different void state categories. The importance is ranked from high to low as follows: the top plate temperature of the foaming process equipment > the isocyanate gun tip pressure of the foaming machine > the temperature of the foaming isocyanate tank; the characteristic judgment rule of the void state is: when the top plate temperature of the foaming process equipment is ≤35.67°C, the isocyanate gun tip pressure of the foaming machine is ≤132.05bar, and the temperature of the foaming isocyanate tank is less than or equal to 18.49°C, the foaming void state of the foaming product is unqualified, and when the foaming process characteristics are at other parameter values, the foaming void state is qualified.
[0052] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should be regarded as within the scope of protection of the present invention.
Claims
1. A data-driven method for analyzing the root causes of foaming voids in foaming products, characterized in that: The following steps are involved: Step 1, collecting raw data of foaming cavity infrared detection and each step of foaming process; Step 2: Perform data preprocessing on the collected raw data, and form a structured foam product foaming big data sample {foaming product ID, process characteristics of each link, foaming cavity state label} according to the timestamp; Step 3, a) dividing the foaming big data samples of structured foaming products into a training set and a test set, and randomly sampling part of the sample data in the training set as a validation set; b) using the remaining data in the training set to build a decision tree training model, using the Gini index as the analysis basis, iteratively selecting the best features and feature value division points until the stop condition is met, and obtaining a mapping relationship model for reflecting the process characteristics of each link and the foaming void state label, and selecting hyperparameters based on the validation set; evaluating the model performance of the training model on the test data set based on the model evaluation index, and obtaining an ideal void state prediction model; c) using the trained decision tree ideal model to predict the foaming product with unknown void state; Step 4: According to the mapping relationship between process characteristics and void status in the established model and the calculation result of the Gini index, the branching process of the decision tree subtree is determined to obtain the importance ranking of process characteristic factors affecting the void status and the characteristic rules that lead to the discrimination of different void status categories.
2. A data-driven foam product foaming void root cause analysis method according to claim 1, characterized in that: The collected data on each link of the foaming process of the foaming products include: the premixing stage, preheating stage, injection stage and foaming stage of the foaming process as well as the environment in which the process is located; the process information characteristics of each link include: foaming gun tip flow, gun tip pressure, foaming process temperature, material tank temperature, foaming injection amount, foaming agent ingredient ratio, preheating temperature, and environmental temperature, humidity and atmospheric pressure.
3. A data-driven foam product foaming void root cause analysis method according to claim 1, characterized in that: The data preprocessing operations include identification, cleaning, deduplication, and standardization.
4. A data-driven foam product foaming void root cause analysis method according to claim 1, characterized in that: In the data integration process, the void state is first determined based on the infrared detection results of the foaming voids, and the labeling is completed. The labeling result is whether the foaming void state of the current foaming product is qualified or unqualified; then, based on the foaming process timestamp and the foaming product ID as the index, a structured foaming product foaming big data sample {foaming product ID, process characteristics of each link, foaming void state label} is established; the foaming void state is used as the subsequent model output response, and other process characteristics are used as model feature inputs.
5. A data-driven foam product foaming void root cause analysis method according to claim 1, characterized in that: The sample size ratio of the training set and the test set fluctuates between 9:1 and 7:3, and the sample size ratio of the validation set in the training set to the training set fluctuates between 1:9 and 1:
8.
6. A data-driven foam product foaming void root cause analysis method according to claim 1, characterized in that: Model hyperparameters include: the maximum number of branches of decision tree leaf nodes, the maximum depth of the tree, the minimum number of samples required for node partitioning, and the maximum number of leaf nodes; hyperparameter search methods include: grid search, random search, or Bayesian search; stopping conditions include: all samples contained in the current node belong to the same category, or the sample set contained in the current node is empty, or the current attribute set is empty, or all samples have the same values on all attributes and cannot be divided; model evaluation indicators include: error rate, accuracy, precision rate, recall rate, F1 value, AUC value, and ROC value, which are used to prune the model according to the evaluation indicator results to prevent overfitting.
7. A data-driven foam product foaming void root cause analysis method according to claim 1, characterized in that: The Gini index calculation formula is: For a given sample set D, its Gini index is defined as: Where Gini represents the Gini index, D represents the current data set, and C k is the subset of samples in D that belong to the kth class; If the sample set D is divided into two sample sets D1 and D2 based on whether a certain feature a takes a certain value, then under the condition of the value of feature a, the Gini index calculation formula of set D is: Wherein, a represents a certain feature of the foaming process characteristics; In the candidate attribute set A, select the attribute that minimizes the Gini index after partitioning: a * =argmin[Gini(D,a)], where a∈A Among them, a * It is used as the optimal partitioning attribute and clarifies the feature value partitioning points. It is also used as the basis for the subsequent quantification and ranking of the importance of feature contribution levels for classification problems. In addition, the generated decision tree branch structure is used to implement feature rules for distinguishing different void state categories.
8. A data-driven foam product foaming void root cause analysis method according to claim 7, characterized in that: By performing weighted cumulative calculation on the Gini coefficient values of different process characteristics at each branch node of the decision tree, the quantitative value of the importance level of the process characteristics contribution to the foaming cavity state classification problem is obtained, and the importance of the process characteristics of each link is further ranked from large to small.
9. A data-driven foam product foaming void root cause analysis system, characterized in that: The system includes a data acquisition module, an integrated processing module, a model building module and a root cause analysis module; Data acquisition module, which collects the original data of foaming cavity infrared detection and each step of foaming process; The integrated processing module performs data preprocessing operations on the collected raw data to construct a structured foam product foaming big data sample {foaming product ID, process characteristics of each link, foaming cavity state label}; Model building module, a) divide the big data samples of structured foaming products into training set and test set, and randomly sample some sample data in the training set as the validation set; b) use the remaining data in the training set to build a decision tree training model, and use the Gini index to iteratively select the best features and division points until the stop condition is met, so as to obtain a mapping relationship model that reflects the process characteristics of each link and the foaming void state label, and select hyperparameters based on the validation set; evaluate the training model based on the model evaluation index, and obtain the ideal model based on the model performance on the test data set; c) use the trained decision tree ideal model to predict the void state of unknown foaming products; The root cause analysis module determines the branching process of the decision tree subtree based on the established model mapping relationship and the Gini index calculation results, obtains the importance ranking of the process characteristic factors affecting the void state and the characteristic rules that lead to the discrimination of different void state categories.