Smart manufacturing is the inevitable trend of technological development of stock companies in the new situation, and it is also an important guarantee for winning the initiative and being invincible in the fierce market competition. The construction of 1580 hot rolling smart workshop is Baosteel's first demonstration smart project as a pilot project of the Ministry of Industry and Information Technology of the People's Republic of China. By introducing and upgrading existing technologies and equipment, Baosteel has taken the step of smart upgrade of the steel industry.
This article will introduce in detail how Baosteel uses the intelligent algorithm based on JMP Pro to predict the natural widening of 1580 hot-rolled finish rolling. It is also a supplement to the content of the speech by Mr. Zhu Jianqin, the chief engineer of digital simulation of Baosteel Equipment Department at 2020 JMP Discovery Summit China .
One Project background — Why need JMP Pro ?
In the construction project of Baosteel's 1580 hot rolling smart workshop, the finishing rolling dynamic width control (DWC) system was used by Primetals Technologies Germany GmbH (PT) to predict the natural width of the finishing rolling and a self-adjusting neural network model Embedded in the L2 system, through the input of the L2 process parameters, the predicted value of the natural width of the finishing rolling is calculated, used to determine the rough rolling target width and self-adjustment after rolling, and transmitted to L1 as a reference value, and the finishing rolling width is realized by L1 This has significantly improved the control accuracy of the exit width of the 1580 hot-rolling finish rolling. The average deviation between the predicted value of the model and the actual value is only 1.38mm (January to September 2019), see Figure 1.
figure 1
However, the precision-rolled natural broadening prediction and self-adjusting neural network model provided by Pratt in Germany is a black-box model. The original code is encrypted and cannot be read or modified, and there is no model structure description. Improvements such as parameter adjustment and model optimization.
Therefore, the team hopes to use JMP statistical analysis of other machine learning, deep learning and other artificial intelligence modeling tools in the product family to develop an intelligent model for the natural width prediction of finishing rolling at Baosteel Co., Ltd. 1580 hot rolling mill. Replace the black box model of German PT company, and make the deviation between the predicted value and actual value of the self-developed model reach the monthly average value ≤ 1.7mm, thus basically reaching the control level of the German black box model.
Based on this, the team began to make further attempts and explorations with the help of predictive modeling tools in JMP Pro.
two The establishment of intelligent forecasting model
1 Data collection and cleaning
The team collected 1580 hot-rolled finish rolling production data from 2019.01 to 2019.09 for nine months, and compiled a SAS program to clean up and integrate the data from January to August 2019. A total of 85,423 strips of effective data were extracted and summarized. Form an Excel data file.
2 Data filtering
01 Use smart algorithms to filter key factors
Apply the Bootstrap forest algorithm in the JMP Pro statistical analysis software to the data set (from January to August 2019, 85423 records, 206 influence factors) for fitting training, and then from the training results according to the actual natural widening The degree of contribution (according to the regression sum of squares), select the influence factors with the sum of squares of variance> 3000, a total of 31, see Figure 2.
figure 2
02 Pratt ( PT ) Inheritance of company experience
The natural widened neural network model provided by the German Pratt (PT) company analyzes the input/output interface of the diagnostic system. It is not difficult to see that the neural network black box model requires 33 components such as chemical composition, finishing width of entrance and exit. Enter the variables, see Figure 3.
image 3
03 Data synthesis
The team compared the key factors selected by the Bootstrap forest algorithm of JMP Pro with the input variables required by Pratt's (PT) neural network model. It can be found that 4 influence factors are the same, and they are pre-calculated precision. Rolling target width and chemical composition silicon (Si), niobium (Nb), etc.
At the same time, based on the production quality analysis experience, it is not difficult to see that there is autocorrelation between the actual rough rolling width, the actual value of the rough rolling half width, the rough rolling target width and other factors, and these variables and the team need to fit the target actual natural width (Spread_ACT) There is also a linear correlation. Using JMP Pro’s multivariate linear correlation analysis tool to analyze these eight influencing factors and the actual values of natural broadening, it can be seen that there is a strong correlation between the eight influencing factors (correlation coefficient r>0.99), and the The correlation of the actual value of the broadening is also relatively strong (r>55%). See Figure 4.
Figure 4
In addition, there are more than 1,000 alloy codes. Although they have an impact on natural broadening and contribute the most, they are too large to fit, so you can use a table method for self-adjustment learning.
Therefore, after discussion, it was finally decided to delete these 9 variables, plus the repeated 4 variables to delete a total of 13 influence factors. In this way, there are 18 remaining influencing factors of 31, plus 33 factors inherited from Pratt's experience, and finally 51 influencing factors are determined.
3 data preparation
In response to the final 51 influencing factors that affect the natural spread of finish rolling, the team collected three months of data from June to August 2019 (a total of 30533 strips), used JMP Pro to read these data, and The actual value of the natural width of the finishing rolling is used as the target value (Y, response variable), 51 influencing factors are used as the fitting data (X, parameters), and 80% of the fitting data set is randomly set as the training set and 20% as the Validation set.
three Smart model training and offline fitting
The team tried to use two intelligent modeling methods in the JMP Pro analysis software: neural network and Bootstrap forest method to train and verify the above data sets, and get the following results:
1 Applied neural network modeling method
Figure 5 Figure 6
It can be seen from Figure 5 and Figure 6 that the model obtained after training and fitting using the neural network algorithm, the coefficient of determination R2 after verification with the verification data is 0.858, the RMSE is 1.506, and the average absolute deviation is 1.16, which theoretically has reached The goals set by the team.
Figure 6 shows the scatter diagram of the predicted values of the three models (from left to right, JMP model, PT model and Baosteel’s existing model) and actual values. It can be seen that the model fitted by JMP has the smallest degree of dispersion.
2 application Bootstrap Forest method modeling method
Figure 7 Figure 8
It can be seen from Figure 7 and Figure 8 that the model obtained after the Bootstrap forest method is trained and fitted has a coefficient of determination R2 of 0.863 and an RMSE of 1.48 after verification with the verification data, which achieves higher prediction accuracy.
Similarly, Figure 8 shows the scatter diagrams of the predicted values of the three models (from left to right, JMP model, PT model, and Baosteel's existing model) and actual values. It can be seen that the model fitted by JMP has the smallest degree of dispersion. )
It can be seen from the application of the above two smart model algorithms that Bootstrap The fitting effect of the forest algorithm is better than that of the neural network, and the prediction accuracy is higher. Both of these modeling methods have reached the goal of the project team.
However, this is only the result of the theoretical analysis of the data. What is the actual application effect is to use the actual control data on the spot to speak. How to apply the models established by these smart algorithms to actual control is really meaningful work.
The model export function of JMP Pro software builds a bridge from theory to practice for the team, and bridges the gap between offline analysis and online application, and makes the ditch into a thoroughfare.
four Online verification of smart models
1 Model export
The JMP Pro software is used to convert these two smart model algorithms into C language codes, thus taking a critical step for the practical application of smart control models in the field.
2 Online verification of models
The C language code converted from the two smart model algorithms is compiled and integrated into the preset process of the 1580 hot-rolling roughing process computer offline analog machine, and then tested using the data in September 2019, that is, the input factor is According to the actual data of strip steel in September, the smart model is used to predict the natural width of the finishing rolling, and then the same standard deviation calculation formula of Pratt PT is used to calculate the standard deviations of the Baosteel model, the PT model and the JMP smart model. See the figure below 9:
Picture 9
It can be seen from Figure 9, From the actual forecast results, JMP The prediction result of the model fitted by the forest method has reached the prediction result of Baosteel’s existing model !
Fives Project summary and experience
By using the intelligent analysis and modeling methods in the JMP Pro statistical analysis software, the team trained, verified and fitted the predicted value, actual value and related parameters of the natural spread of 1580 hot-rolled finish rolling, and obtained the neural network and The two intelligent control models of Bootstrap Forest, their theoretical prediction accuracy and verification accuracy completely reach the accuracy of the neural network model of German Pratt company.
At the same time, with the help of the model release function of the JMP Pro software, the team obtained its C language code, and carried out the actual test of the preset process on the analog machine of the on-site program control computer, and achieved very good results. The one-step online production test and verification laid a solid foundation.
In addition, during the analysis and modeling process, the team also summarized the following experiences for your reference:
- The time span of training materials should not be too long, 3 to 4 months is appropriate;
- The amount of data for training verification should not be too large, and 30,000 to 40,000 records are appropriate;
- If equipped with a self-adjusting learning function, the prediction accuracy can be higher.
It can be seen that with the introduction and advancement of the Made in China 2025 strategy, the manufacturing industry is gradually transforming and upgrading from manufacturing to "smart manufacturing", and it is imperative for smart manufacturing to empower future steel mills. The trial and exploration of Baosteel's 1580 hot-rolling finish rolling natural widening project provides a good reference and basis for the smart manufacturing road of Baosteel's other product lines. Scientific analysis tools will surely become indispensable for the manufacturing industry to move towards poetry and distance. Powerful engine.
