A Hybrid System for the Design and Processing of C-Mn and Microalloyed Steels

P. A. Manohar, R. J. Dippenaar and S. S. Shivathaya*

BHP Institute for Steel Processing and Products,

University of Wollongong, Northfields Avenue,

Wollongong, NSW - 2522, Australia.

*Hawker de Havilland Ltd., 361 Milperra Road,

Bankstown, NSW - 2200, Australia.

ABSTRACT

A novel hybrid approach is proposed in this paper that combines knowledge bases along with mathematical modelling approaches to generate and evaluate the alternative target compositions for steelmaking, which meet the customer requirements. The methodology developed is applicable for C - Mn, Nb- and Ti- microalloyed steels. The system consists of two modules. The first module uses both mathematical (iterative) and knowledge-based approaches to generate a list of alternative target compositions for steel making. The target compositions are then evaluated by using the second module that consists of mathematical modelling approach to calculate the microstructural evolution as a function of steel composition and process route. The mechanical properties of the steel products are then computed based on the microstructural parameters using empirical relationships contained within the structure - property model. This method enables more realistic assessment of the designed compositions. The system is expected to assist the product development metallurgists in the selection of appropriate target composition for steelmaking and for hot rolling process optimisation.

INTRODUCTION

The challenges for iron and steel industry are many fold and complex, eg. being more flexible and responsive, less capital intensive, energy efficient and “earth friendly”. Non-ferrous metals and non-metallic materials have provided continuous competition to steels as alternative engineering materials. To deal with these challenges and for efficient management of the uncertainties involved, it has become imperative to apply artificial intelligence (AI) techniques in the manufacture of iron and steel. In the past three decades a number of expert systems have been developed around the world for a more efficient solution of problems in diagnostic, design, planning, scheduling, process control, and quality control (1). Several expert systems have been developed and utilised in almost all aspects of iron and steelmaking such as primary operations (eg. blast furnace, sinter plants), steelmaking (eg. BOS monitoring and control), rolling operations (eg. plate mill, hot strip mill), energy and utilities (eg. electric power system), systems engineering, quality management (eg. defect analysis in continuous casting), research, planning, sales price forecasting, and purchase (2). Application of a knowledge-based approach to steel composition design has been considered by (1, 3 - 5). However, the development of such expert systems is a complex task because the material design process is ill structured, difficult to systematise and involves a large number of rules. In addition, linear relationships do not exist between compositional and process parameters and product properties and as such the knowledge of steel composition design is largely intuitive and heuristic.

On the other hand, the hot rolling of steels has been investigated intensively and a number of mathematical models and computer systems have been developed. Several advantages gained through the use of mathematical modelling have been reported which include the following:

improving the efficiency of mill trials to establish optimum process conditions, controlled rolling schedules and accelerated cooling (6),

prediction of microstructure and mechanical properties during rolling (7),

development of new steel grades and rolling processes (8, 9),

increase productivity and quality, reduce manufacturing cost through the use as an off-line prediction, on-line prediction, on-line control or off-line alloy and process design tool (10, 11),

ability for flexible manufacturing (12),

control of size, shape, quality and stability of steel products, more responsive for product development (13),

betterment of understanding of the processes (14),

a useful “what - if” tool which provides directions for further fundamental research along with problem investigation, schedule development, design or redesign of mill configuration and enhancement of understanding (15).

In the current work, a new integrated approach is proposed to build an expert system which combines the above two approaches to generate and evaluate the alternative compositions for steelmaking that meet the customer’s requirements. The expert system consists of two modules. The first module uses both mathematical (iterative) and knowledge-based approaches and utilises interview as well as non-interview techniques for knowledge elicitation (KEL). KEL is also characterised by a three-character codification scheme to record customer’s special requirements. The codification scheme is coupled with a decision table-based knowledge representation tool “TABLEAUX” for incorporation within knowledge-based systems. The expert system generates a list of alternative target compositions, which may meet the property requirements. The compositions are then evaluated by using the second module that consists of mathematical modelling approach. The module calculates the microstructural evolution as a function of steel composition and known values of process variables such as pass temperature, strain, strain rate, interpass time and plate cooling rate during the hot rolling of C-Mn and Nb- and Ti- microalloyed steels. The predicted microstructure is used as a basis for the subsequent estimation of the mechanical properties of the steel products using empirical relationships, thus enabling more realistic assessment of the designed compositions. The expert system is developed in C / C++ language on an IBM PC in a windows environment. User interface is developed utilising a commercial package, ‘PROTOGEN+’, to make the expert system user friendly. The expert system is expected to assist the product development metallurgists in the selection of appropriate steelmaking target composition and for hot rolling process optimisation.

KNOWLEDGE ELICITATION

The process for knowledge elicitation (KEL) adopted in this work has been reported in detail elsewhere (1, 16 - 17), however a brief summary is given here. The KEL is characterised by a three-character codification scheme having a hybrid structure to codify all the customer’s special requirements based on the initial structured and unstructured interviews. The customer special requirement codes (CSRCs) are given by the equation:

Customer Special Requirement Code =XiYjZk

The first character in the code is Xi called the major group code, which is the ith property of a steel grade (eg. tensile strength, yield strength, elongation etc.). The second character in the CSRC is called the subgroup code and it represents the jth type of steel (eg. structural, pressure vessel, line pipe steel etc.). Zk is the value code which represents the kth value of the ith property of the jth type of steel. Zk has a hierarchical structure while Xi and Yjare chain type structures. The chain type structure facilitates vertical (depth first) search while the hierarchical structure assists the horizontal search. A total of 238 328 CSRCs are possible using this codification scheme. The significance of using these CSRCs is that the knowledge representation becomes simple, time efficient and memory efficient (i.e. requires less storage space).

Knowledge representation is a key development stage in KEL. The acquisition and organization of knowledge for incorporation within the expert system being reported here has been achieved through a decision table based tool ‘TABLEAUX’ developed at BHP Steel Company, Port Kembla, Australia (18). A decision table is a collection of rules. Each rule is represented as a row made up of conditions and actions within the table, an example of such rows is given in Figure 1. Figure 1 illustrates the knowledge representation in TABLEAUX along with the use of codification scheme. Codification is achieved by combining columns 2 - 4 in Fig. 1(a) in to a three-letter code as shown in Fig. 1(b). The conditions are to the left of bold line in Fig. 1 (a) and Fig. 1(b) while the actions are to the right of the bold line. For any rule to be evaluated true, it is necessary that all conditions within the rule match the values in their corresponding cells. For example, Rule 2 in Figure 1 can be stated in natural language as follows:

IF

the steel type is structuralAND

the test type is reduction in area in transverse direction (RAZ)AND

the minimum value of RAZ is 15%

THEN

maximum Sulphur content is 0.008%AND

maximum Hydrogen content is 19 ppmAND

maximum Calcium content is 0.01%AND

critical caster alignment is A1AND

electro-magnetic stirring code is E4AND

sulphur print requirement is 1

Rule / Steel Type / Test Type / Value / Max.
S / Max. H / Max. Ca / Align-ment / EMS / S Print
Rule 1 / Struct-ural / RAZ / 25% Min. / 0.005% / 19 ppm / 0.01% / A1 / E4 / 1
Rule 2 / Struct-ural / RAZ / 15% Min. / 0.008% / 19 ppm / 0.01% / A1 / E4 / 1

(a)

Rule / SR Code / Max.
S / Max.
H / Max.
Ca / Align-ment / EMS / S Print
Rule 1 / 211 / 0.005% / 19 ppm / 0.01% / A1 / E4 / 1
Rule 2 / 212 / 0.008% / 19 ppm / 0.01% / A1 / E4 / 1

(b)

Figure 1: Knowledge representation (a) before codification, and (b) after codification.

Knowledge rules such as those given above have been determined through interview as well as non-interview techniques of KEL. The knowledge rules also pertain to both composition and steel processing. The knowledge bases developed in this work contain a large number of such rules that elucidate complex interrelationships between steel composition, processing variables and product properties. The details of the knowledge bases and their interconnection in the present expert system are discussed in the following sections.

EXPERT SYSTEM DEVELOPMENT

The expert system consists of four knowledge bases and a steel-processing module. The inputs and outputs of the system are shown schematically in Figure 2. The knowledge base KB I consists of information on properties and composition corresponding to relevant material standards. The material standards include Australian standards and other overseas standards transformed into a form, which is similar to the Australian standards. Customer special requirements are also included in KB I. Based on the customer special requirements, the composition and mechanical properties from the existing steels need to be modified. This is achieved through the knowledge rules contained in the second knowledge base, KB II.

Input information about the end use of the steel or the intended application of the steel along with information in KB I and KB II dictates a set of rules regarding elements to be included in the target composition and the basic process route to be followed. The process routes could be hot rolled, controlled rolled, or normalised. These rules are included in the third knowledge base, KBIII that calculates the upper and lower limit values of the elements to be included in the target composition. Some values of the elements in the target composition, in spite of being within the range of values obtained through KB III, are not feasible due to practical difficulties faced by either the plate mill or the slab caster. In addition, based on the end use and the mechanical properties required, certain strategies need to be adopted in the design of the target compositions. Such strategies impose further restrictions on the target composition values. Thus the rules regarding process limitations and design strategies are contained in the fourth knowledge base, KB IV. Some examples of such rules are:

increment in C (C)= 0.005%,

Nb = 0.001%,



Cu:Ni  2.0,

Ti:N  3.42,

Mn:C 3.

The output from KB IV and the process details given in KB III are combined in steel processing module, which calculates the metallurgical structure evolution as a function of composition and process sequence. Details of the steel processing module are given in the following section.

Expert System

Figure 2: Input, processing and output of the expert system.

STEEL PROCESSING MODULE

The flow chart for the steel processing module is given in Figure 3. Mathematical modelling of microstructural evolution during hot rolling of steels has received a great deal of attention over the past two decades and a number of models which describe metallurgical phenomena during steel processing have been published for different steel compositions (eg. C-Mn, Nb-/Ti-/Nb-Ti/Nb-V microalloyed steels) and a variety of steel processing routes (eg. conventional, conventional controlled rolling, recrystallization controlled rolling, hot direct rolling etc.). These models have been reviewed in (11, 19). The mathematical models employed in the current work for calculating the microstructural evolution in Nb- microalloyed steels are given in Table 1.

START

INPUT

RECRYSTALLIZATION

MODEL

PRECIPITATION

MODEL

RollingNo

Finished?

Yes

PHASE TRANSFORMATION

MODEL

STRUCTURE - PROPERTY

MODEL

END

Figure 3: Flow chart for the steel processing module.

Table 1: Mathematical models describing the microstructural evolution

during the hot rolling of Nb- microalloyed steels.

Parameter / Model / Reference
pass strain
 / 1.155 ln(ho/hf) / (15)
pass strain rate
/ VR/ / (15)
time for 5% recrystallization
t0.05 / 6.75x10-20xdo2x-4 x exp(300000/RT)xexp{((2.75x105/T)-185)x[Nb]} / (20)
time for 5% precipitation of
Nb(C, N)
t0.05p / 3x10-6x[Nb]-1x-1x Z-0.5 x exp(270000/RT)x
exp{(2.5x1010)/(T3(lnKs)2)} / (20)
Nb supersaturation ratio
Ks / {[Nb]+([C]+0.86[N])} /
10(2.26-6770/T) / (20)
time for 50% recrystallization t0.5 / 4.92x10-17x -2x -0.33 x do x exp(338000/RT) / (21)
volume fraction recrystallized X / 1-exp(-0.693(t/t0.5)2) / (22)
recrystallized grain size
drex / 1.1xdo0.67/67 (X0
0.5xdo0.67/67 (X<0 / (23)
(9)
Zener - Hollomon parameter
Z / / (20)
time for 95% recrystallization t0.95 / 7.64 x t0.05 / (22)
grain growth during
interpass time ‘t’
df / df8.75 - do8.75= 2.6x1028x
exp(-437000/RT)xteff ;
teff = t - t0.95 / (24)
average austenite grain size when X < 0.9
(partially recrystallized austenite)
/ eff = pass + 
 = const. x previousx(1-X)
const. = 1 if X < 0.1; 0.5 if X  0.1;
= X1.33xdrex+(1-X)2do / (9)
(15)

ho = original slab thickness (mm), hf = final slab thickness (mm), T = pass temperature (K), t = interpass time (s), VR = peripheral roll speed (mm/s), RR = roll radius (mm), R = gas constant (8.31 J/mol.K), do = initial grain size (m).

The mechanical properties for each steelmaking target composition are calculated based on the output from KB IV and the steel processing module. Empirical models derived from the statistical data are utilised for this process. The empirical models are characterised by an error of about  20 MPa in the prediction of tensile strength and upper yield strength. A factor of safety of 40 MPa is added to the required values of tensile and upper yield strength while comparing with the corresponding computed values. Thus the final target composition list is generated which has alternative target compositions that are feasible for any inquiry or order.

SUMMARY

The new integrated expert system for steel composition design proposed in this work combines both mathematical modelling and knowledge-based approaches. Mathematical modelling enables iterations involving enormous computations while the knowledge-based approach enables utilisation of the expert as well as the heuristic knowledge from a group of experts to successfully determine the steelmaking target compositions. The procedure involved in this approach is to identify the possible customer requirements with regard to composition, mechanical properties and testing requirements, then to codify them, coupled with processing schedules used in the industry and finally to direct the KEL to acquire knowledge to deal with these special customer requirements. The quality of the output of this expert system depends mainly on the quality of the rules in knowledge bases and mathematical models. As the knowledge base grows richer by experience and the mathematical models refined further through research, it is always possible to incorporate more rules into knowledge bases to improve the output of the expert system. The expert system is expected to assist metallurgists to choose an existing composition or to design a new steel composition so that the customer requirements are satisfied in an economical way. The prototype material design expert system has been fully implemented by developing a software module for generating alternative steelmaking target compositions that are practically feasible for the slab caster and plate mill. Implementation of the process optimisation module is currently under development.

ACKNOWLEDGEMENTS

The authors acknowledge the financial support extended by the Australian Research Council and BHP Steel Company, Australia during the course of this work.

REFERENCES

1. Shivathaya S. S. (1997). Material design in steelmaking utilising mathematical modelling, knowledge-based and fuzzy logic approaches. Ph. D. Thesis, University of Wollongong, Australia.

2. Noderer K. and Henein H. (1996). A survey on the use of expert systems in the iron and steel industry. Artificial Intelligence for Ironmaking and Steelmaking, ISS, USA, 175 - 179.

3. Shivathaya S. S. and Fang X. D. (1996). A hybrid material design and evaluation system for steelmaking. ISIJ International, 36, 424 - 432.

4. Watanabe M., Iwata Y., Obama N., Shirahata K., Suematsu K., Meiga T. and Yamane H. (1991). Development of shape steel quality design expert system. Nippon Steel Technical Report, No. 53, 29 - 33.

5. Vasko F. J., Wolf F. E. and Stott K. L. (1989). A set covering approach to metallurgical grade assignment. European Journal of Operational Research, 38, 27 - 34.

6. Abe T., Honda T., Ishizaki S., Wada H., Shikanai N. and Okita T. (1990). Application of computer modelling of thermo-mechanical processing on steel plate production. Proc. int. symp. on ‘Mathematical Modelling of Hot Rolling of Steel’, ed. by Yue S., Hamilton, Canada, 66 - 75.

7. Choquet P., Fabregue P., Giusti J., Chamont B., Penzant J. N. and Blanchet F. (1990). Modelling of forces, structure and final properties during the hot rolling process on the hot strip mill. Proc. int. symp. on ‘Mathematical Modelling of Hot Rolling of Steel’, ed. by Yue S., Hamilton, Canada, 34 - 43.

8. Hodgson P. D. and Gibbs R. K. (1990). A mathematical model to predict the final properties of hot rolled C-Mn and microalloyed steels. Proc. int. symp. on Mathematical Modelling of Hot Rolling of Steel, ed. by Yue S., Hamilton, Canada, 76 - 85.

9. Laasraoui A. and Jonas J. J. (1991). Prediction of temperature distribution, flow stress and microstructure during multipass hot rolling of steel plate and strip. ISIJ International, 31, 95 - 105.

10. Kwon O., Lee K. J., Lee J. K. and Kang K. B. (1995). Modelling of austenite evolution and transformation for MA strips. Proc. int. conf. Microalloying ‘95, ed. by Korchynski M.et al., ISS, Pittsburgh, USA, 251 - 261.

11. Kwon O. (1992). A technology for the prediction and control of microstructural changes and mechanical properties in steel. ISIJ International, 32, 350 - 358.