Abstract JSW Steel Ltd produces different finished product namely hot rolled strip, cold rolled strip, and long products. The objective of this work was to produce regression based models to capture and understand the relationship between the variables and mechanical properties of steel in different product forms. There are many factors that affect the mechanical properties in the process. The models were used to interpret the trends by altering one variable and keeping the others constant. Such studies are often not possible to conduct experimentally. In addition, the predictions of the regression models were validated with the metallographic studies and the plant data. Specific analysis has been carried out for the similar composition (IS: 11513 D) steel being processed through the hot strip mill (HSM), cold rolling mill (CRM) and wire rod mill (WRM). Impact of process parameters on the mechanical properties has been analysed using these models and validated with microstructure and plant data. Results to date indicate that the accuracy of the model is good and is suitable for the evaluation of new steel grades and the development of optimised thermo-mechanical processing routes. Keywords: MLR (Multi Linear Regression), HSM (Hot Strip Mill), CRM (Cold Rolling Mill) and WRM (Wire Rod Mill). INTRODUCTION JSW is an integrated steel plant that has a Hot Strip Mill (HSM), Cold Rolling Mill (CRM) and Long Product Mill (LPM) with a capacity of 10 million tonnes per annum. It can roll a wide range of steels comprising low carbon, tube, structural, API line pipe, micro alloyed, low Al, low Si and several medium carbon steels in three process routes. Its products are hot rolled coils, cold rolled coils and long products, which are rolled in the respective mills. A data-based approach proves good, especially for industrial systems because of their non-complexity. Often not even process experts are aware of all the relationships of their systems and are able to understand every behavior. This paper constitutes mechanical properties modelling for three different process routes. Regression has been carried out for all three process route variables to predict the mechanical properties of these products. The main feature of the regression is the establishment of linear relationships between data whi ch makes thi s techni que very attracti ve for understanding the direct impact of process variables. In the presented application area, one big challenge is the fact that such processes typically are rich in data but poor in terms of information, since measurements are made in a few operating points and an arbitrary excitation is not possible owing to cost concerns. Identifying the working window for quality improvement can be made easy with the help of such models. MODEL FORMULATION The model has been formulated by using several explanatory data like chemistry, temperature and process variables to predict the response variable like mechanical property. The goal of multiple linear regressions (MLR) is to model the relationship between the explanatory and response variables. The model for MLR, given observations, is: y i = B 0 + B 1 x i1 + B 2 x i2 + ... + B p x ip + E i Where i = 1, 2, ..., n The relationship between y i and x i is then estimated by carrying out a simple linear regression analysis. Least squares criterion has been used to estimate the equations, so that it is possible to minimise the sum of squares of the differences between the actual and predicted values for each observation in the sample, that is, minimise 6 E i . Although there are other ways of estimating the parameters in the regression model, the least squares criterion has several desirable statistical properties, most notable of which is that the estimates are maximum likelihood if the residuals E i are normally distributed. Fig.1. shows the schematic model formulation. Prediction of Mechanical Property of Low Carbon Steel Processed in Different Rolling Routes through Regression Pradeep Agarwal, R. Madhusudhan, L. Naidu, Abhay Shrivastava, Vinoo D. S. and S. Manjini JSW Steel Ltd, Vijayanagar Works, INDIA VOL. 7 NO. 2 JANUARY, 2013 STEEL TECH 54 Fig. 1: Schematic model formulation The data for the analysis work was collected from the respective rolling mills. Number of test data for each mill is mentioned in Table 1. Chemical and dimensional range of input variables are given in Table 2. Inputs Outputs (explanatory variables) (response) variables) Chemistry Model calculations Temperatures and Process variables Empirical relations Dimensions Type of mill Yield Strength Ultimate Tensile Strength Elongation Table 1: Number of test data with each production unit Type of production unit Number of test data used for analysis Hot Strip Mill (HSM) 18300 Cold Rolling Mill (CRM) 18800 Wire Rod Mill (WRM) 7100 EMPIRICAL RELATION DEVELOPMENT The empirical relations were developed by running the model for all the mills products to predict mechanical properties which are given below. The data include the following variables (the symbols of the variables are shown in brackets). Tensile strength (MPa) [TS], yield strength (MPa) [YS], % Elongation [%EL] and % Reduction in area [%RA] Element concentrations from actual ladle analysis (weight %) [C], [Mn], [Si], [Nb], [V], [N], [Ti], [Cr], [Cu], [Ni], [P], [S], [B], [Al] Table 2: Range of data used with each production unit Chemistry HSM CRM WRM (wt %) Min. Max. Min. Max. Min. Max. C 0.001 0.64 0.001 0.1 0.06 0.8 Mn 0.05 1.73 0.05 1 0.3 0.9 Si 0.002 0.9 0.003 0.9 0.01 0.35 P 0.002 0.1 0.002 0.025 0.004 0.03 S 0.002 0.04 0.003 0.025 0.001 0.03 Al 0.002 0.1 0.025 0.065 0.002 0.06 N 0 0.009 0 0.006 0 0.006 Ti 0 0.1 0 0.04 B 0 0.005 0 0.005 Cr 0 1.2 Cu 0 0.4 Nb 0 0.09 Ni 0 0.45 V 0 0.16 Dimension 1.6 26 0.3 3.2 5.5 22 (mm) STEEL TECH 55 VOL. 7 NO. 2 JANUARY, 2013 Finishing temperature [FT], Coiling temperature [CT] for hot rolled strip Shape parameters [thickness], [width] of strip, [diameter] of wire-rod and gauge length [GL] of tensile sample. Processing parameters for CRM i.e. total reduction in cold compact mill [% reduction] and skin pass mill elongation [%SPM elongation]. Annealing parameters for CRM i.e. outer temperature of coil in batch annealing furnace [case temperature], inner ring temperature of coil [core temperature], total heating time and holding time in furnace [heating hood time] It may be noted that all the temperatures are in [C], dimensions are in [mm] and time is in [sec]. DEVELOPMENT OF EMPIRICAL RELATIONS FOR HSM MECHANICAL PROPERTIES Hot strip mill processing involves a series of thickness reduction step at high temperature for input slab to make strip. First the slab is heated in reheating furnace (1190 1250 C) to make it thermally homogenised and dissolve most of the carbo-nitrate precipitates. Then the rough rolling is carried in reversing rougher mi l l that breaks cast mi crostructure and form recrystallised grains of austenite. The high temperature of roughing process (1040 1200 C) allows higher strai n i n passes wi thout maki ng materi al work harden [1]. The roughing mill exit temperature is typically set to 1000-1120C depending on requirement of finishing temperature or microstructural requirement to improve the properties. The finish rolling is carried out in 6 stand four high finishing mill. The typical finish rolling temperature is set close to Ar3 temperature (820-910C) and coiling temperature is set as per the end microstructural requirement (540-700C). Final mechanical properties evolve mainly owing to phase transformation behavior, grain size of ferrite and precipitation strengthening. The empirical relation developed for UTS, YS and % EL for hot rolling mill (HSM) is shown below: R 2 (UTS) = 0.95 UTS = 449 + 620(C) + 75(Mn) + 490(P) 16(S) + 92(Si) + 2.1(Al) + 856(N) 3109(B) + 43(Cr) + 249(Cu) +2866(Nb) 68(Ni) -142(Ti) + 211(V) - 0.09(FT) - 0.1(CT) -2.3(thicknes) R 2 (YS) = 0.87 YS = 460 + 364(C) + 64(Mn) + 327(P) 7.5(S) + 48(Si) + 22(Al) + 324(N) 4305(B) + 35(Cr) + 475(Cu) + 3862(Nb) - 408(Ni) 236(Ti) + 264(V) 0.14(FT) 0.12(CT) 4.8(thickness) R 2 (EL) = 0.81 EL = 45.4 49(C) - 3(Mn) - 33(P) - 54(S) - 2(Si) - 1.7(Al) + 220(N) - 360(B) - 3(Cr) -29(Cu) - 219(Nb) + 23(Ni) + 33(Ti) + 96(V) + 0.0026(FT) + 0.0143 (CT) + 0.24 (thickness) 0.223(GL) DEVELOPMENT OF EMPIRICAL RELATIONS FOR CRM MECHANICAL PROPERTIES The hot rolled strip is used as a input material for cold rolling. First the surface scale is removed through continuous pickling line operation. The cold thickness reduction take place in 6-Hi twin stand reversible compact cold rolling mill. The total reduction typically varies between 70-90% of input thickness. The material become work harden owing to cold rolling. Batch annealing treatment (below Ar1 temperature) is given to get the required microstructure and properties. Finally skin pass rolling is performed as per customer requirement. The mechanical properties are mainly dependent on mi crostructural changes duri ng anneal i ng and precipitation strengthening. The empirical relation developed for UTS, YS and % EL for cold rolling mill (CRM) is shown below: R 2 (UTS) = 0.84 UTS = 152 + 121(C) + 188(Mn) + 667(P) 245(S) + 2.75(Si) - 167(Al) + 2401(N) 1.77(thickness) 0.003(width) + 0.16(% reduction) -0.14(% SPM elongation) -0.001 (case temperature) + 0.13(core temperature) -0.38(total heating time) 2.1(heating hood time) R 2 (YS) = 0.79 YS = 122 + 11(C) + 228(Mn) + 320(P) 215(S) - 6.75(Si) - 216(Al) + 4451(N) - 1.9(thickness) 0.003(width) - 0.37(% reduction) -0.17(% SPM elongation) -0.001 (case temperature) + 0.02(core temperature) -0.05(total heating time) 2.8(heating hood time) R 2 (EL) = 0.67 EL = 49.7 - 4(C) - 21(Mn) -39(P) + 15(S) + 0.25(Si) + 9.8(Al) - 272(N) + 0.22(thickness) + 0.001(width) - 0.01(% reduction) + 0.01(% SPM elongation) -0.01 (case temperature) + 0.01(core temperature) + 0.08(total heating time) + 0.31(heating hood time) DEVELOPMENT OF EMPIRICAL RELATIONS FOR LPM MECHANICAL PROPERTIES: Continuously cast billets (165*165 mm 2 ) are the input material for making wires and rods. The cross section VOL. 7 NO. 2 JANUARY, 2013 STEEL TECH 56 WRM ASTM Grain size no. = 9.25, SD = 0.009 UTS = 379MPa, %RA = 73, HRB = 49 Product Diameter = 5.5mm Finishing Temperature = 910C Coiling Temperature = 660C of square billet is reduced through series of horizontal and vertical reductions in 30 stands mill. First the billet is heated (1000-1230C) in furnace to make it thermally homogenise. The billet then passes through breakdown mill, roughing mill, finishing mill, no twist mill and reducing/sizing mill to attain the final shape. The end rolling temperature is typically set above Ar3 temperature (850- 930C) to complete rolling in austenitic region. Different mode of cooling can be applied to get the desired microstructure. Finally the product is coiled in vertical pallet reform station. The empirical relation developed for UTS, YS and % EL for long product mill (LPM) is shown below: R 2 (UTS) = 0.83 UTS = 301+ 512(C) + 91(Mn) + 32(Si) + 169(P) + 62(S) 88(Al) + 696(N) 0.6(diameter) R 2 (%RA) = 0.72 %RA = 83 58(C) 4.4(Mn) + 15.7(Si) 14(P) - 18(S) + 13(Al) 42(N) - 0.7(diameter) The coefficient of reliability (R 2 ) was fairly good for MLR. Micro-alloying, finishing temperature and coiling temperature was found to be mainly affecting the mechanical properties for hot rolled products. Annealing parameters were mainly affecting the properties of cold rolled products. Fairly good relations could be achieved with just chemistry and shape correlation for the mechanical properties in wire rod products. It may be owing to less number of total grades with similar process variability in wire rod category. The detail effect of each element and process variable has already been published repeatedly [1-4] hence not included in this paper. COMPARATIVE ANALYSIS OF THREE MILLS PRODUCTS One single composition (IS: 11513 D) has been selected from the products of three mills namely HSM, CRM and WRM as shown in Table 3. Full process analysis, microstructure evolution and mechanical properties had been recorded and analysed. The results are given in Table 3. The predicted mechanical properties in Table 4 are fairly close to the tested mechanical properties in Fig. 2. The variation in mechanical properties in different processes is directly related to how the microstructure evolved during thermo-mechanical processing along with phase transformation. Details are given in Table 4. Table 3: Chemical composition (IS: 11513 D) Element C Mn Si (max) P (max) S (max) Al (max) N (max) Wt % 0.05-0.06 0.3-0.4 0.05 0.01 0.01 0.05 0.005 Fig. 2: Comparison of microstructure and mechanical properties of similar steel composition being process through different product line HSM ASTM Grain size no. = 9.13, SD = 0.009 UTS = 365 MPa, YS = 294 MPa, %EL = 39 Thickness = 3.2mm Finishing Temperature = 847C Coiling Temperature = 598 C CRM ASTM Grain size no. = 7.87, SD = 0.16 UTS = 320MPa, YS = 190 MPa, %El = 42.6 Thickness = 1.2mm Annealing Temperature = 700C, 590C at core Heating Time = 17.4 Hr, Holding Time = 2 Hr Water Cooling start temperature = 380C STEEL TECH 57 VOL. 7 NO. 2 JANUARY, 2013 Table 4: Predicted mechanical properties based on process parameters Predicted HSM CRM WRM properties UTS (Mpa) 379 303 387 YS (Mpa) 305 179 % Elongation 35 42.9 % RA 73 In Fig. 2(a), the microstructure had evolved after the processing in HSM. It mainly contains the polygonal ferrite with discreet distribution of cementite particles at the grain boundaries [1] . After the finish rolling at 847C, the complete structure was austenite with elongated grains and stored deformation [2] . This austenite rapidly recrystallised in to fine equiaxed structure [2] . Phase transformation take place in runout table where the average cooling rate was found to be 30C/sec. Fine equiaxed austenite yielded in to fine equiaxed ferrite after phase transformation. This phase transformation completes before coiling at 598C. After coiling some amount of grain growth and cementite coarsening is expected for this low carbon steel [3] . In Fig. 2(b), the microstructure had evolved after processing in CRM. It mainly contains coarse ferrite with aspect ratio 2.4 and fine distribution of cementite within it. The pancake structure is owing to cold working; only sub-grains are eliminated, complete recrystallisation did not take place [4] . After the cold reduction the structure was full hard owing to strain hardening. This material was then annealed in batch annealing furnaces, for the targeted core temperature of 590C, while the gas temperature was maintained at 700C. It took 17.4 Hr for heating the coil to this temperature. Heating hood was placed in the furnace for 2 Hr; finally it was air cooled to 380C and then water cooled to room temperature. Annealing had removed most of the strain hardening and final structure is soft ferrite. In Fig. 2(c), the microstructure had evolved after processing in WRM. It mainly contains polygonal ferrite and some cementite. After the finish rolling at 910C, material was cooled with force air. The average cooling rate was found to be 25C/sec i n the phase transformation temperature range. Similar to HSM line, recrystallised austenite yielded in fine polygonal ferrite. Difference in mechanical properties comes owing to the difference in cooling rate after coiling. The cooling rate was found to be much higher for WRM products than in the HSM coils. CONCLUSION Empirical relations through MLR have been developed with fairly good accuracy. It can be directly used to know the mechanical properties in advance. Comparison on microstructure between the products of three mills reveals the difference in thermo mechani cal processi ng of these l i nes. Fi nal mechanical properties have been understood in terms of microstructure evolution during and after phase transformation. The potential application of the model can be identified as, Prediction of mechanical properties in advance. Optimisation of rolling temperatures for improved microstructure. New product development and improvement in processing parameters of existing grades. This model can be further useful for the beginners to understand the effect of chemistry and rolling on the final microstructure and properties. REFERENCES 1. R.W.K. Honeycombe, Steels: Microstructure and Properties, Edward Arnold, London, 1981. 2. Yu-Tuo Zhang, Dian-Zhong Li Modelling of austenite decomposition in plain carbon steels during hot rolling Journal of Materials Processing Technology 171 (2006), pp. 175179 3. Pradeep Agarwal, Manohar Namani, Madhusudana R, and Madhu Ranjan, Optimisation of Rolling Parameters for Improved Grain Structure in Drawing Grade Steel Steel Tech. Vol. 4, No. 2, Jan 2010, pp. 33-37 4. Robert W. Cahn and Peter Haasen; Physical Metallurgy Volume 1; 1996 (Fourth revised edition)