| Home | My Account | Post Article | Upload Showcase | Site Shortcuts | Feedback | About us |

  Home > Consulting > Level 2 Model >

Seven Tech
Pat. Warn
Pat. Level2
Prj cases
Model Sys
New L2
Ind. Softw.
Key Prjs
Smart Equip
Model Prjs
Li Batt Ind.
Steel Ind.


Defect Warn

Equip Intelli

Domain Sale


- Class Time
- Classes
- Resources
- Lrn Method
- Status Need
- Metaverse
- Li Batt Tech
- Equip Softw.
- Intelli System
Shape, Property, Productivity

NISCO Plate/Coil Mill Level 2 Force Model Improvements

Bingji (Benjamin) Li
Metal Pass LLC
Pittsburgh, PA, USA

Pengju Zhu
Nanjing Iron & Steel Co. (NISCO)
Nanjing, Jiangsu, PR China

Daoyuan Wang
Nanjing Iron & Steel Co. (NISCO)
Nanjing, Jiangsu, PR China

Key words: Plate coil mill, Level 2, Force prediction, Flow stress learning, Metallurgical effect


Level 2 force prediction serves as basis for draft schedule generation and initial roll gap setup before AGC adjustment. This paper covers Level 2 force model improvements in NISCO plate/coil steckle mill, especially for continuously increasing steel grades, and for automatic design of flow stress coefficients based on chemical composition, slab and product mix, and rolling process including thermomechanical rolling. The guided two-parameter learning, by designing flow stress coefficients with metallurgical effects integrated and learning logic issues resolved, proves to be an easy-to-apply and very accurate solution. Force prediction accuracy is thus significantly increased. Further development fields are also listed.

Following topics will be covered:

1.1 Development Background

1.2 Force Prediction as Basis for Flat Rolling Draft Scheduling (
1.3 Mill Production and Product Quality Issues

2.1 Level 2 Learning Logics Issues and the Solution
2.2 The Guided Two-Parameter Fitting (FIT2G)

2.3 Data Collection and History Log Processing (
2.4 Steel Grade Consolidation
2.5 Model Grade Design

2.6 Grade Families Design (
2.7 Automatic Design of Flow Stress Coefficients

2.8 Grade File Generation (
2.9 Resume Pass Model Improvement
2.10 Miscellaneous Improvements
2.11 Source Code Modification

2.12 Solution to Continued Grade Increase and Missing Grade File (
2.13 Onsite Testing
2.14 Force Prediction Accuracy




1.1 Development Background

Nanjing Iron and Steel Co. (NISCO) is one of the China's major steel producers, with operations ranging from mining, ironmaking, steelmaking, and steel rolling to produce flat, wire rod and special sections, etc. For steel plate production, it operates a plate mill, plate/coil (steckle) mill and a heavy wide plate mill (in construction).

Level 2 force prediction improvement in Metal Pass LLC initially started in USA for Evraz Oregon Steel (EOS) [1]. In the second half of 2006, EOS Level 2 models frequently encountered shape defects, especially for the hard and thin products. Analysis indicated that the Level 2 models had significant force errors, in the related passes [1]. The high force prediction error in the Level 2 model caused unreasonable draft schedules that led to bad finish shape. The early stage of the work was filled with confusion because many phenomenons were contradictory to the traditional rolling theory. Later, a new theory system was established [2]. On this basis, Metal Pass quickly proposed the development on next-generation Level 2 system with metallurgical principle, intelligent learning and advanced software engineering (e.g. uninterrupted upgrade) [3]. In the Level 2 model, two fundamental issues, the force prediction accuracy, and the draft schedule logic to create rolling schedule based on predicted force, both are tightly linked with, or better to say, interacte with, metallurgical issues. For plate rolling or strip rolling, Level 2 cannot create draft schedule without force prediction. If the force prediction is inaccurate, the draft schedule will not be appropriate unless the scheduling logic is defective. In this case, there would be waste in mill capacity and damage in product quality.

This paper focuses on one of the two most fundamental issues of the Level 2 draft scheduling, the force prediction, with example of NISCO, in China. NISCO Plate & Coil Mill is newer than EOS plate mill, but it belongs to the earliest successful suppliers of high-quality X80 steels in China. This mill was also built by Tippins and uses Tippins Level 2 system. Like every mill of this type, its Level 2 model has force prediction problem due to several Level 2 logic issues, together with the limitation of adaptive learning and the lack of metallurgical consideration which are the common weaknesses in most Level 2 packages available in the market. Many sources of errors cannot be simply removed by blind adaptive learning even with neural network optimization.

Today, it is very common to add micro-alloys into steel. Addition of Nb, V and Ti often increases recrystallization temperature by about 100C, while use of Cr, Mo and Cu, etc. often pushes up the recrystallization temperature by about 50C. Therefore, addition of micro-alloys makes it harder for steel to complete recrystallization, and consequently, significant strain could be retained from previous passes. Force prediction without consideration of those issues would leave the predicted result inaccurate and thus downgrade the draft schedule quality and even cause equipment damage. The controlled rolling would significantly add severity to this problem.

NISCO has pushed very hard in controlled rolling. This creates special issues in the Level 2 model generation for good force prediction. Besides, in serving wide range of industries, its steel grades expand very quickly. This again, increases the difficulty in model generation. Metal Pass usually performs pre-design for flow stress coefficients for every model grade in every of the three temperature regions, in order to apply its so-called Guided Two Parameter Fitting (FIT2G). FIT2G is very simple and effective in considering metallurgical effects and in fixing common learning logic weaknesses in the force prediction. In today technical level, direct modeling for microstructure evolution during rolling and interpass cooling would be a dead end, especially for the online model in which calculation speed is critical. However, this pre-design faces challenge when number of steel grades continues to grow. For this reason, in this project, Metal Pass fully integrated its data, rolling process models and technical skills, and developed a procedure to determine the flow stress coefficients automatically.

<To Be Continued>

<Previous> | <Next>

|  Part1  |  Part2  |  Part3  |  Part4  |  Part5  |  Part6  |  Part7  |  Part8  |  Part9  | 


  | Private Policy | Terms & Conditions | About Us | AdvertisePartnerInvestorSponsorlistings |  
Copyright: 2022 Metal Pass LLC. All right reserved