High-standard bearing steel quality prediction model based on machine learning

Abstract

The invention discloses a high-standard bearing steel quality prediction model based on machine learning. The high-standard bearing steel quality prediction model comprises the following steps: 1) data acquisition: collecting on-site key process and component data having potential influence on the qualification rate and defects of high-standard bearing steel; 2) data preprocessing: processing the collected various process and component characteristic variables; 3) data division: dividing the data into a training set and a test set according to a proportion; 4) constructing a model: training a deep neural network model constructed by using Python through the training set; 5) performing model evaluation: performing evaluation comparison on the trained model through a test set, and taking the trained integrated learning model as a prediction model; and 6) quality prediction: importing the data of the steel to be tested into the trained model to obtain a prediction result. According to the method, a perfect enterprise digital process and a component acquisition system are constructed, the high-standard bearing steel quality is accurately predicted by utilizing a machine learning technology, and the accuracy rates of water immersion flaw detection qualification prediction and crack defect prediction are both greater than or equal to 80%.

Description

Technical Field

[0001] The present invention relates to the field of metallurgical intelligent manufacturing technology, and in particular to a high-standard bearing steel quality prediction model based on machine learning. Background Art

[0002] High-standard bearing steel is a steel specially designed for manufacturing bearings, with high purity, high hardness and wear resistance, good hardenability, suitable toughness and plasticity, corrosion resistance, and precise size and shape control. These characteristics ensure the reliability, durability and precise operation of bearings, and play a decisive role in improving the overall performance of mechanical equipment and reducing maintenance.

[0003] In the steel industry, quality inspection is a crucial link, which is directly related to the performance and safety of the product. Traditional quality inspection methods mainly include automatic flaw detection by machine equipment and manual flaw detection. However, both methods have certain limitations. Although automatic flaw detection by machine can improve the detection efficiency, due to the limitations of the device itself and the detection accuracy of the equipment, many microscopic inclusions are often unable to be found, so manual re-inspection is required to ensure the accuracy of the test results. Manual flaw detection is affected by the subjectivity of the operator, and is prone to low detection efficiency due to high fatigue, and it is difficult to meet the needs of large-scale production.

[0004] With the rise of industrial big data, a large amount of production data can be collected and analyzed, providing new opportunities for quality prediction and control. In recent years, machine learning technology has been widely used in many fields due to its powerful data processing capabilities, including but not limited to bearing quality prediction, fault diagnosis, etc. Especially in the steel industry, the research on steel quality prediction using machine learning technology has gradually increased.

[0005] However, there is currently a lack of effective technology that can effectively identify steels with fewer defects by comparing the microstructure images of steels and accurately predict the quality of steels, which needs to be developed in a targeted manner. Summary of the invention

[0006] Purpose of the invention: In view of the shortcomings and defects of the prior art, the present invention provides a high-standard bearing steel quality prediction model based on machine learning, constructs a complete enterprise digital process and component collection system, and uses machine learning technology to accurately predict the quality of high-standard bearing steel. The accuracy rate of water immersion flaw detection prediction is ≥80%, and the accuracy rate of crack defect prediction is ≥80%.

[0007] Technical solution: A high-standard bearing steel quality prediction model based on machine learning of the present invention is characterized by: comprising the following specific methods to obtain:

[0008] 1) Data collection: Analyze the actual production conditions of each process, collect key process and component data that have potential impact on the qualified rate and defects of high-standard bearing steel, including water immersion flaw detection qualified rate and crack length data;

[0009] 2) Data preprocessing: According to actual needs, various process and component characteristic variables collected in step 1) are processed, which specifically includes cleaning up wrong values ​​and meaningless data in the data and filling in missing values;

[0010] 3) Data division: The cleaned data in step 2) is divided into a training set for model training and a test set for model testing in a ratio of 7:3, 8:2 or 3:1;

[0011] 4) Model building: training a deep neural network model built using Python using the training set to obtain a trained model;

[0012] 5) Model evaluation: The trained model is evaluated and compared using the test set. If the evaluation and comparison are qualified, the trained integrated learning model is used as the prediction model;

[0013] 6) Quality prediction: import the production data and composition data of the steel to be tested into the trained model to obtain the corresponding quality prediction results.

[0014] The data collected in step 1) include electric furnace parameters, refining furnace parameters, vacuum furnace parameters, slag sample parameters, environmental factor parameters and equipment status parameters.

[0015] Among them, in the step 2), abnormal, unconventional and meaningless data are screened out; the feature average value is used to fill the feature missing value, and the feature average value is used to replace the feature abnormal value.

[0016] Wherein, constructing the model in step 4) specifically includes the following steps:

[0017] 1) Use Python to build a deep neural network model and use the SHAP algorithm;

[0018] 2) Write and debug the input layer, hidden layer and output layer of the neural network model; the input layer uses the cleaned feature variables; the output layer uses the water immersion flaw detection pass rate and crack length as output; the hidden layer is initially set to three layers;

[0019] 3) Optimize the performance and generalization ability of the model by adjusting the hyperparameters of the neural network model, including the number of layers, number of nodes, and activation function;

[0020] 4) Design the graphical user interface of the model, including parameter input and prediction result output.

[0021] The output end of the design model includes the water immersion flaw detection pass rate and the crack length.

[0022] Among them, the deep neural network model is replaced by a multi-layer perceptron regression algorithm model, a random forest regression algorithm model, an extreme gradient boosting regression algorithm model, and a linear regression algorithm model according to actual data conditions.

[0023] Among them, the model evaluation in step 5) specifically includes the following steps.

[0024] 1) Use the training set to train the model, adopt appropriate optimization algorithms and loss functions, iteratively update the model parameters, and gradually converge the model; use the test set to evaluate the generalization ability and prediction performance of the model, and adjust the model and optimize the parameters according to the verification results;

[0025] 2) If the test result is too low, optimize the model again and re-divide the training set and test set for training and testing;

[0026] 3) The accuracy of model prediction is continuously improved after multiple rounds of cycles until the prediction accuracy is ≥ 90% after the introduction of real-time field data testing.

[0027] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0028] The present invention uses industrial data and machine learning technology to accurately predict the quality of high-standard bearing steel, including the qualified status of water immersion flaw detection and the size of crack defects. The application of this model can help manufacturers to achieve precise control and optimize production processes, improve product quality, and reduce production costs. At the same time, through early prediction, companies can adjust production plans in a timely manner to avoid unnecessary waste and production interruptions. This will greatly improve production efficiency, reduce quality risks, and further promote the intelligent development of the industrial field.

[0029] The model system of the present invention is used to build a complete enterprise digital process and component collection system, and the machine learning technology is used to accurately predict the quality of high-standard bearing steel, with a water immersion flaw detection qualified prediction accuracy of ≥ 80%, and a crack defect prediction accuracy of ≥ 80%. At the same time, the key process and component parameters are sorted to help production companies achieve precise control and optimize production processes, improve product quality, and reduce production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific implementation methods.

[0032] The high-standard bearing steel quality prediction model based on machine learning of the present invention is obtained by the following specific method:

[0033] 1) Data collection: Analyze the actual production situation of each process, collect key process and component data that have potential impact on the qualified rate and defects of high-standard bearing steel on site, including water immersion flaw detection qualified rate and crack length data.

[0034] 1.1) Analyze the actual production conditions of each process. The collected data mainly includes four types of process information: electric furnace parameters, refining furnace parameters, vacuum furnace parameters, and slag sample parameters;

[0035] 1.2) Data expansion: In addition to the existing key process and composition data, you can also consider collecting other data that may be related to the slab flaw detection pass rate, such as environmental factors, equipment status, etc.

[0036] 2) Data preprocessing: According to actual needs, the various process and component characteristic variables collected in step 1) are processed, which specifically includes cleaning up erroneous and meaningless data in the data and filling in missing values.

[0037] 2.1) Filter out abnormal, unconventional, and meaningless data according to actual needs;

[0038] 2.2) Use the feature mean to fill the missing values ​​of the feature and use the feature mean to replace the feature outliers.

[0039] 3) Data division: The cleaned data in step 2) is divided into a training set for model training and a test set for model testing in a ratio of 7:3, 8:2 or 3:1.

[0040] 4) Model building: The deep neural network model is trained using Python through the training set to obtain a trained model.

[0041] 4.1) Use Python to build a deep neural network model and use the SHAP algorithm;

[0042] 4.2) Write and debug the input layer, hidden layer and output layer of the neural network model; the input layer uses the cleaned feature variables; the output layer uses the water immersion flaw detection pass rate and crack length as output; the hidden layer is initially set to three layers;

[0043] 4.3) Optimize the performance and generalization ability of the model by adjusting the hyperparameters of the neural network model, such as the number of layers, number of nodes, and activation function;

[0044] 4.4) Graphical user interface of the design model, including parameter input terminal and prediction result output terminal. The output terminal of the design model includes water immersion flaw detection pass rate and crack length.

[0045] For deep neural network models, you can use multi-layer perceptron regression algorithm model, random forest regression algorithm model, extreme gradient boosting regression algorithm model, and linear regression algorithm model according to actual data conditions. Select the appropriate model based on the actual training results.

[0046] 5) Model evaluation: The trained model is evaluated and compared through the test set. If the evaluation and comparison are qualified, the trained integrated learning model is used as the prediction model.

[0047] 5.1) Use the training set to train the model, adopt appropriate optimization algorithms and loss functions, iteratively update the model parameters, and gradually converge the model; use the test set to evaluate the generalization ability and prediction performance of the model, and adjust the model and optimize the parameters according to the verification results;

[0048] 5.2) If the test result is low, optimize the model again and re-divide the training set and test set for training and testing;

[0049] 5.3) After multiple rounds of cycles, the accuracy of model prediction is continuously improved until the prediction accuracy is ≥ 90% after the introduction of real-time field data testing.

[0050] 6) Quality prediction: import the production data and composition data of the steel to be tested into the trained model to obtain the corresponding quality prediction results.

[0051] The present invention adopts advanced machine learning technology, which can more accurately predict the quality of high-standard bearing steel, and adjust the production plan in time to avoid unnecessary waste and production interruptions; use machine learning algorithms to process large amounts of complex production data, such as chemical composition and process parameters, to enhance in-depth understanding of the data and explore the relationship between data equipment systems; by analyzing the prediction results, the chemical composition and process parameters in the production process can be optimized to produce higher quality high-standard bearing steel.

[0052] Example:

[0053] The present invention is applied to the actual production of high-standard bearing steel in a steel plant. After data preprocessing, the parameters of the electric furnace, refining furnace, vacuum furnace and slag sample are determined as shown in Table 1:

[0054] Table 1: Model input data categories

[0055]

[0056] In the prediction of 300 heats of steel to be tested, the pass rate of water immersion flaw detection and the hit rate of crack length are shown in Table 2:

[0057] Table 2: 300 heat prediction results

[0058] Number of test furnaces Number of hit furnaces Water immersion flaw detection pass rate hit rate 300 277 92.3％ Number of test furnaces Number of hit furnaces Crack length hit rate 300 263 87.6％

Claims

1. A high-standard bearing steel quality prediction model based on machine learning, characterized by: The following specific methods are included: 1) Data collection: Analyze the actual production conditions of each process, collect key process and component data that have potential impact on the qualified rate and defects of high-standard bearing steel, including water immersion flaw detection qualified rate and crack length data; 2) Data preprocessing: According to actual needs, various process and component characteristic variables collected in step 1) are processed, which specifically includes cleaning up wrong values ​​and meaningless data in the data and filling in missing values; 3) Data division: The cleaned data in step 2) is divided into a training set for model training and a test set for model testing in a ratio of 7:3, 8:2 or 3:1; 4) Model building: training a deep neural network model built using Python using the training set to obtain a trained model; 5) Model evaluation: The trained model is evaluated and compared using the test set. If the evaluation and comparison are qualified, the trained integrated learning model is used as the prediction model; 6) Quality prediction: import the production data and composition data of the steel to be tested into the trained model to obtain the corresponding quality prediction results.

2. The high-standard bearing steel quality prediction model based on machine learning according to claim 1 is characterized in that: The data collected in step 1) include electric furnace parameters, refining furnace parameters, vacuum furnace parameters, slag sample parameters, environmental factor parameters and equipment status parameters.

3. The high-standard bearing steel quality prediction model based on machine learning according to claim 1 is characterized in that: In the step 2), abnormal, unconventional, and meaningless data are screened out; the feature average value is used to fill the feature missing value, and the feature average value is used to replace the feature abnormal value.

4. The high-standard bearing steel quality prediction model based on machine learning according to claim 1 is characterized in that: The step 4) of constructing the model specifically includes the following steps: 1) Use Python to build a deep neural network model and use the SHAP algorithm; 2) Write and debug the input layer, hidden layer and output layer of the neural network model; the input layer uses the cleaned feature variables; the output layer uses the water immersion flaw detection pass rate and crack length as output; the hidden layer is initially set to three layers; 3) Optimize the performance and generalization ability of the model by adjusting the hyperparameters of the neural network model, including the number of layers, number of nodes, and activation function; 4) Design the graphical user interface of the model, including parameter input and prediction result output.

5. The high-standard bearing steel quality prediction model based on machine learning according to claim 4 is characterized in that: The output of the design model includes the water immersion flaw detection pass rate and the crack length.

6. The high-standard bearing steel quality prediction model based on machine learning according to claim 4 is characterized in that: The deep neural network model can be replaced with a multi-layer perceptron regression algorithm model, a random forest regression algorithm model, an extreme gradient boosting regression algorithm model, or a linear regression algorithm model according to actual data conditions.

7. The high-standard bearing steel quality prediction model based on machine learning according to claim 1 is characterized by: The model evaluation in step 5) specifically includes the following steps. 1) Use the training set to train the model, adopt appropriate optimization algorithms and loss functions, iteratively update the model parameters, and gradually converge the model; use the test set to evaluate the generalization ability and prediction performance of the model, and adjust the model and optimize the parameters according to the verification results; 2) If the test result is too low, optimize the model again and re-divide the training set and test set for training and testing; 3) The accuracy of model prediction is continuously improved after multiple rounds of cycles until the prediction accuracy is ≥ 90% after the introduction of real-time field data testing.

Images