A Level 2 system consists of the function blocks to perform process modeling (Level 2 Model) and those to collect and manage data. In developing a new Level 2 system, full metallurgical models should be applied besides the traditional mechanical models. In the following, the general Level 2 system architecture will be outlined first, and then the metallurgical models for the Level 2 will be discussed.

The newly developed Level 2 system should have following features:

1. Full metallurgical principles integrated. For example, a rolling mill Level 2 system may be integrated with over 100 rolling process models, an expert system, and advanced intelligent learning. Microstructure simulation would be performed pass by pass to determine parameters such as retained strain, flow stress, grain size, grain shape, phase proportion, etc. The modeling, such as microstructure simulation, could be integrated with the AI learning techniques and with an expert system, and be continuously learnt through history data. Finite element method, especially a simplified one such as that developed by Dr. Benjamin Li (Metal Data), would be used to determine, e.g. roll deflection and local draft over width. Hybrid systems would be established by combining the AI learning techniques with the empirical models, so that the models are not black boxes (as many neural networks are) and will be continuously improved.

2. Uninterrupted upgrade. The Level 2 system could be upgraded frequently using vendor-supplied or user-developed components, without system shutdown. It would be fully modularized and fully object-oriented. The system would facilitate uninterrupted upgrade in three levels: service, component and DLL. A service usually consists of multiple components (including software applications) and is developed following certain IT standards (e.g., SOA, COM+). A component may be created from multiples classes (DLLs). A DLL is created from a single class, which may call other classes following object-oriented methodology. An upgrade in the DLL level may be done, for example, by simply replacing an old DLL with the new one, or by adding new DLLs to the existing Level 2 system. Therefore, with the changing industry practice, the Level 2 vendor (or the third party) could supply new DLLs, Components or services to the user. There would be, therefore, no need to retire an old system and to buy a new one in, say, every ten years. The upgrade cost in this way could be minimal.

3. State-of-the-art software engineering technologies. Fully object-oriented programming technique would be combined with the interactive relationship of the mill process models. SOA (Service-Oriented Architecture) could be used to integrate various applications and components, which may be initially designed for various platforms. The source code is expected to be concise, easy to understand and easy to maintain. As to the architecture, it may be a four-tier system, consisting of Operator Interface (HMI, tier 1), Level 2 Management System (tier 2), Level 2 Model System (tier 3), and Database Management System (DBMS, tier 4). The tier 2 and tier 3 could either reside in a single server or be separated in two or more servers. Due to the large number of model calculations (microstructure, FEM, Neural Network, etc.), separate servers for the tier 3 would be preferred.

Level 2 Model Architecture

For each process (rolling mill, reheating or controlled cooling), the newly developed Level 2 model should include intelligent learning system. Numerical analysis tools (FEM, FDM, etc.) may also be integrated depending on the application case, e.g. FDM for thermal prediction, and FEM for both thermal and mechanical analyses. Level 2 system of any process should have interface with other Level 2 systems.

• Process models and related metallurgical ones, with also the interfacing architecture with other Level 2 model system, including numerical tools (for e.g. temperature analysis for reheating and cooling, and roll deflection prediction for rolling).

• Expert system, which consists of logics, data and influence factors on the mechanical, thermal and metallurgical parameters depending on rolling and thermal processes. Past mill-experiences should be programmed as a portion of the expert system.

• Intelligent Learning, based on rolling process models, in which the neural network provides the correcting factors for the model coefficients, and fuzzy logic rules and expert system provide guidelines (upper and lower boundaries, etc.) for the learning.

See detailed model system architectures for major processes:

• Metallurgical Models for Rolling

• Metallurgical Models for Reheating

• Metallurgical Models for Controlled Cooling