TATA Steel

CASE STUDYProject Leads to $1.65 MillionIn Annual SavingsIMPROVEMENT T ata Steel’s domestic and international market share for galvanized coils,NEEDED TO MEET especially from international markets such as China and Iraq, rose fromRISING DEMAND 23% to 32% during 2003. The Tata location in Jamshedpur, India, wasFOR GALVANIZED not able to meet rising demand.COILS. Tata Steel’s galvanized coils are manufactured in two lines known as continu-By Soumish Dev, ous galvanizing lines one and two (CGL1 and CGL2). To meet the risingP&O Nedlloyd demand and increased sales volume, we decided to initiate a project to increase the speed factor at both lines, starting at CGL1. The project charter set kickoff for March 7, 2003. The data collected for December 2002 to February 2003 revealed the following: • Current average speed factor: 0.874. • Current standard deviation of the speed factor: 0.1391. • Minimum speed factor: 0.077. • Maximum speed factor: 1.780. The data revealed an approximate loss of 10,500 tons of galvanized coils per year caused by a low speed factor. This meant an opportunity loss of about $1.33 million per year, depending on the market price of the galvanized coils. At CGL1, all coil thicknesses are based on the market requirement. The improvement team decided to concentrate on thinner sections (coils having thick- ness less than 0.63 mm) because the rising demand for steel was more for the thin- ner sections. Thinner sections are high-end products yielding greater profit mar- gins. We also observed the thinner section coils had a lower speed factor. We also decided to maintain the current quality level so the existing high quality of the galvanized coils would be maintained. Management set the lower and upper specification limits for the speed factor as 0.95 and 1.2, respectively. We decided not to increase the speed factor beyond 1.2 because doing so would increase the zinc alloy consumption of the coils being galvanized, which would then result in additional cost. Calculating the Speed Factor The supplier’s rated capacity of the furnace at CGL1 is 22.2 tons per hour (tph), and the following calculations were based on this fact: Mass flow rate = density x volume/time = density x area x velocity = density x thickness (or section) x width of the coil x velocity = density x 0.38mm x 1,030mm x velocity (standard being taken as 038 mm and 1,030 mm). The allowable mass flow rate given the capacity of the furnace = 22.2 tph (standard). Therefore, estimated velocity = (mass flow rate)/(density x 0.38 mm x 1,030 mm). The standard density of steel was estimated at 7.85 x 103 kg/m3 (standard), assuming the density is homogeneous over the coil. Therefore, estimated veloc- ity = 22.2tph/(7.35 x 103 kg/m3 x 0.38 mm x 1,030 mm) = 120 meters per I I 37S I X S I G M A F O R U M M A G A Z I N E NOVEMBER 2004

Project Leads to $1.65 Million in Annual Savingsminute (approximately) = rated speed of CGL1. (entry section), bridle 3B (process section) and bridle Thus, estimated time taken by a coil at CGL1 = 5B (exit section).length of the coil/estimated velocity. Because we knew the radius of the roll (r), the lin- To determine the actual time taken by a coil, three ear velocity of the roll was calculated asspeed master encoders at three sections measured the linear velocity = w * r.axial velocity (w) of the rolls in CGL1: bridle 1D Assuming there was no slippage of the sheet overFigure 1. Measurement System Analysis Gage name: Gage repeatablility and reproducibility (gage R&R) for measurement Date of study:Percentage Components of variation Percentage Reported by: 100 contribution Tolerance: 50 Percentage study variation Miscellaneous: 0 Gage R&R Repeat Reproduction Part-to-part By part 100 90 80 70 60 50 40 30 Part 1 2 3 4 5 6 7 8 9 10 R chart by operator By operatorSample range 1.0 Abhishek Mohit Upper control 100 limit (UCL) = 0.4901 90 80 0.5 Lower control 70 60 limit (LCL) = 0 50 0.0 40 30 Operator Abhishek Mohit Xbar chart by operator Operator x part interaction 100 OperatorSample mean 100 Mohit Mean = 62.73 90 Abhishek 90 Abhishek 80 Mohit 80 UCL = 63.01 Average 70 70 60 50 60 LCL = 62.44 40 50 30 40 Part 1 2 3 4 5 6 7 8 9 10 30 Rule of thumb for measurement system analysis: < 2%: accept measurement system. 2% – 8%: business call. > 10%: reject measurement system.Figure 2. Data Collection TabulationClarify data collection plan Operational definition Sampling Type: discrete Unit of How it will be Who HowMeasure (Y) or continuous Associates many? measurement measured Time period in different What WhereSpeed factor Continuous shifts 2,481 Ratio Calculated as a December 2002 Actual Taken total ratio by measuring to March 2003 time from the data the designed time taken encoder points taken by a coil and by a in bridle the actual time coil 3B taken by a coil38 I IN O V E M B E R 2 0 0 4 WWW.ASQ.ORG

Project Leads to $1.65 Million in Annual SavingsFigure 3A. Speed Factor Data Set Figure 4A. Speed Factor Performance For December 2002 Variable: speed factor Mean: 0.900241 Sigma: .076916 1.4 11111 1 11 Upper control Nonnormal fit; Skewness: 0.298190 Kurtosis: 0.128980 1.2 1 11 11 1 limit = 1.046 Specifications: Lower specification limit = 0.950000; 1 111 11 Nominal = 1.00000; Upper specification limit = 1.20000 1.0 Lower specification limit Individual value Mean = 0.8866 0.8 -3.s Nominal +3.s 800 0.6 1 1 1 1 1111 11 111 1 Lower control 700 1 limit = 0.7273 1 0.4 11 1 1 11 1 1Frequency 600 11 1 1 500 0.2 1 11 1 1 1 1 400 0.0 1 300 0 100 200 300 400 500 600 700 200 Observation number 100 0 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 1.05 1.10 1.15 1.20 1.25Figure 3B. Speed Factor Data Set Figure 4B. Speed Factor Performance For January 2003 Probability-probability plot for variable: speed factor 11 1 Upper control Distribution: normal 11 1 limit = 1.062 1.0Observed cumulative distribution function Mean: 0.90024; Standard deviation: 0.07692 Mean = 0.8376 Individual value 1.0 0.5 1 1 1 1 1 111 1 1 1 Lower control 0.9 1 11 11 1 1111 limit = 0.6134 0.8 111111 1 0.7 1 1 11 0.6 11 1 11 111 1 0.5 1 1 11 0.4 0.3 11 1 1 1 11 0.2 0.1 1 0.0 0.0 1 0.0 0 100 200 300 400 500 600 700 800 Observation number 0.2 0.4 0.6 0.8 1.0 Theoretical cumulative distribution function Figure 4C. Speed Factor Performance For February 2003the rolls, the linear velocity of the roll is equal to the Individual value 2 11 1actual linear velocity of the coil. 1 Upper control Thus, actual time taken by a coil at CGL1 = length 1111 limit = 1.105of the coil/actual velocity. The speed factor = 1designed time taken by a coil/actual time taken by a Mean = 0.8617coil. 1Measurement System Analysis 111111 1 11 11 1 11 1 1 111111 11 1 11 Lower control 1 1 11 1 1 11 1 1 limit = 0.6185 Measurement system analysis results are shown in 1 1 1Figure 1. Based on some rules of thumb regarding 1the total percentage contribution of gage repeata- 1 1bility and reproducibility (R&R), the measurement 0 1 11 1 0 100 200 300 400 500 600 700 Observation number I I 39S I X S I G M A F O R U M M A G A Z I N E NOVEMBER 2004

Project Leads to $1.65 Million in Annual SavingsFigure 5A. Tonnage Distribution Chart Tonnage chart Weight of coils in tons 7,000 6,000 5,000 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.63 4,000 583.69 1,567.36 5,402.36 6,771.53 6,273.49 3,442.44 5,491.54 734.85 2,119.52 1,330.78 3,000 2,000 Section in millimeters 1,000 0 TonnageFigure 5B. Average Speed Factor and Standard DeviationSpeed factor Distribution of speed factor over section 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.63 Section in millimeters Average speed factor Standard deviation of speed factorFigure 6. Process Diagram Machine Adjustment of lib gap Man Pressure control valve Frequent Air knife Malfunction Timely • Fear of side tracking (of thinner sections) roll change Improper weld response • Fear of unknown factorsLife of rolls parameters Sequencing of sensors Unavailability of Interlock skilled operators Pot rolls settings problems Fear of operators Nonrotation Alignment failures Welding Lack of positional trainingSmoothness Furnace Condition of Weld wheel change Poor or delayed operationand weld parts Insufficient knowledge Steering about interlocksnonalignment of Problem of Instrument Unable to control side trackinghot bridle roll side tracking malfunction Shortage of Ink jet manpower printers Design/location issue (absenteeism) Machine faults Continuous galvanizing No start up/stoppage during breakdown Unavailability of coils/cranes/ line one maintenance and shut down maintenance operator/business rules speed factor of 0.87 Nozzle chokage Carryover Calorific value/pressure of carbon monoxide gas Chromating Scheduling Off gauge Generation of Unavailability of cranes Incoming baby coils Exit sequence off gauge Entry sequence Alkali Shape not proper Drier carr yover Cold rolled coil qualitySquindle Air knife equipment healthiness is bad roll Bath chemistry Sur face Metallurgical properties Alkali Improper zinc coating defectscleaning Method Malfunctioning, calibration of coating gauge, Material Physical defects control valves, pressure transducers The side tracking seems to be a major concern for low speed factor40 I IN O V E M B E R 2 0 0 4 WWW.ASQ.ORG

Project Leads to $1.65 Million in Annual SavingsFigure 7. Side Tracking Pareto Chart Pareto chart 1,400 120% 1,200 100% 1,000 80% 60% 800 40% 600 20% 400 0% 200 0Number of observations PFortetCqEeoulqnCeduFSsiiionrulQtpWnroLDodulkennonlelmraSNralifaoendeiloetczoldtAddenllulzrlyteFfwelSckeUrssotetolpppppadnsCohShrfluirrrrrticdtueopaasnooooomiaroRstdcDcrckhbbbbibctaallllmlhtdkyeatkiat-oiiarlreeeeaeceulogpeunrnnroalmymlmmggdpenyeskemdp StGWarabuorgnSoegltSlrivchrpaouoirtlitbdaafrttsoeiieiewooankdnknn Cumulative percentage of number of observationsNote: Cushion factor relates to the excessiverelaxed speed of the line due to the fear Reasonsfactor of the operators of side tracking.system was accepted. However, a data collection plan Figure 8. Side Tracking Control-Impact Matrixwas developed before we moved on to the analyzephase. The data collection plan is tabulated in Figure Low High Ease of control2 (p. 38). As you can see, the total number of data Low Highpoints over the three month period was 2,481. • Quench tank rolls. • Camber in cold rolled shape. • Misalignment of • Gage variation. The process capability study revealed that • Campaign sizealthough the speed factor is continuous data, it does furnace rolls.not follow a normal distribution. We conducted an (raw material planning).Anderson Darling test1 and later a Kolmogorov- • Cold rolled roughness. • Furnace authorization.Smirnov2 test to find out the best fit distribution. • Entry looper position. • Jet cool fan operation. • Alkali carryover (*). As you can see from Figures 3A and 3B (p. 39), • Steering unit response. • Sink roll problem (*).the speed factor data set did not follow normal dis- • Deflector roll. • Tensions fluctuations (*).tribution. We thus could calculate the process capa- • Furnace temperature settings.bility through defects per million opportunities • Welding operation (*). • Misalignment of bridle two rolls. • Entry looper rail alignment. • Hot bridle roll alignment.Figure 9. Pareto Chart of Quality Function Deployment for Side Tracking Entry looper rail alignment 50 100 150 200 250 300 350 Hot bridle roll alignment Pressure difference in snubber roll Pot tension Looper tension Squeeze roll pressure after alkali sectionSqueeze roll pressure before hot rinse section Squeeze roll pressure after hot rinse section Furnace tension Drier temperature Horizontal misalignment of sink roll Vertical misalignment of sink roll Bush clearance Pressure difference in pinch roll Alignment of clamps Squeeze roll pressure before alkali section 0Note: The quality function deployment template was developed in-house using Microsoft Excel and some macros based on certain assumptions,such as no benchmarking, no competitor figures and no interrelationship between the how’s or X’s). I I 41S I X S I G M A F O R U M M A G A Z I N E NOVEMBER 2004

Project Leads to $1.65 Million in Annual SavingsFigure 10. Reasons for Side Tracking Categorization of failures 100%Number of observations 80% 60% 40% 20% 0% 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.63 0.2 Section in millimeters Cold rolled shape DelayCushion Side tracking Soft cycle Rated Alkali carry Quality problemEquipment problem Star t-up Furnace pressure fluctuation Sink roll problem Pot tension fluctuationsFold mark New sink roll Cold rolled shortage Welder problem(DPMO) methodology. The short-term sigma level of Figure 11. Pareto Chart of T-Testthe current process was 1.5. The detailed calculation ofthe process capability in the DPMO methodology was: Pareto chart of t-values for coefficients; df = 31 Variable: speed factor • Project Y = speed factor. Length 4.747 • Lower specification limit = 0.95; upper specifica- Thickness 4.5546 tion limit = 1.20; target = 1.00. 3.559116 Weight 2.62082 • Total number of observations of speed factor Pressure diffrerence in snubber roll (units) = 2,481. .880864 Looper tension .7254955 • Total number of observations of speed factor less Pot tension .711023 than 0.95 = 1,872. .7055769 Squeeze roll pressure after alkali section .7028721 • Total number of observations of speed factor Squeeze roll pressure before hot rinse section .4947776 more than 1.2 = 13. .1382579 Squeeze roll pressure after hot rinse section • Total number of defects = 1,872 + 13 = 1,885. Furnace tension • Total number of opportunities = 1. Drier temperature (Because only speed factor was measured through theaxial velocity of the encoders, the number of opportu- p = 0.05nity/chances of not meeting the critical to quality is 1). T-value (for coefficient; absolute value) • Thus, DPMO = [1,885/1 x 2,481)] x 106 = 759,774. • The yield of speed factor = 24.02%. From the tables, The yield of speed factor refers to the fact that only • Long-term sigma level = 0. 24.02% of the coils that ran through CGL1 during the three-month period had a speed factor between 0.95 • Short-term sigma level = 1.5. and 1.2. Any coil that had a speed factor less than 0.95 or more than 1.20 did not qualify as high end product • The classical yield = 1 – D/U = 1 – 1,885/2,481 = and thus was treated as scrap. 1 – 0.7598 = 0.2402. (Classical yield is yield that doesn’t account for any rework of scrap.) Charts were drawn to see the performance of the speed factor over a period of time. The charts are42 I IN O V E M B E R 2 0 0 4 WWW.ASQ.ORG

Project Leads to $1.65 Million in Annual Savingsshown in Figures 4A, 4B and 4C Figure 12. Response Surface Drawing(p. 39). A tonnage distributionchart (Figure 5A, p. 40) was also 3-D surface plot (data stratification 16v x 44c)drawn to observe each section of Speed factor = distance weighted least squarescoils. The coil categories werebased on the section thicknessof the coils. Figure 5B (p. 40)also shows the average speed 2.2factor and its standard deviation 2.0across different sections of the 1.8 1.8coils. These were all drawn to 1.6 1.4allow us to understand the dif- 1.4 1ferent behavior of the speed fac- Speed factor 1.2 0.6tor across different sections. 1.0 0.2 0.8 Once the different behavior of 0.6the project Y was understood and 0.4its process capability calculated, 0.2we focused on the different 0.0enablers (or the X parameters).To enlist the different enablers, 0.605.T6h00ic.k5n05e.s5s00.405.400.305.300.205.20 3.0 3.2Pre3s.4sur3e.6diff3e.r8enc4e.0in 4.2 4.4 rollwe drew a detailed flowchart for snubberthe process, starting with the coilscheduling on the line and end-ing when the galvanized coil Analysis of variance for speed factor (coded units)came out of the line. Degrees of Sequential Adjusted Adjusted The fishbone diagram (Figure6, p. 40) revealed side tracking as Source freedom sum of squares sum of squares mean of squares F-value P-value Main effects 4 0.88879 0.537257 0.134314 7E + 04 0.003the most prominent concern. Two-way interactions 6 0.39641 0.615606 0.102601 5E + 04 0.003Side tracking is a phenomenon in 0.44386 0.443859 0.110965 6E + 04 0.003which the coil comes out of the Three-interactions 4 0.00000 0.000002 0.000002 0.00000 0.000002 0.000002 Residual error 1 Pure error 1track or the line at high speed Total 15 1.72906and causes serious damage to themachines and sometimes injuredworkers. Moreover, because the coil does not meet the ed two recommendations from the QFD during aspecific quality level, it is treated as scrap. monthly shutdown of the plant in June 2003: To drill down further to the reasons for side tracking, 1. We fixed the entry looper rail alignment by fixingwe developed a Pareto chart (Figure 7, p. 41), a control- the distance between the rails to 2,200 mm, withimpact matrix (Figure 8, p. 41) and a Pareto chart based a tolerance less than 0.5 mm.in quality function deployment (QFD, Figure 9, p. 41).Figure 10 shows the different reasons for side tracking 2. We fixed the hot bridle roll alignment set by align-across the different section of the coils. We implement- ing roll number one, which was skewed by 2 mm.Figure 13. Pilot Project Solutions Actions recommended Implementation plan Details of tools used Quality function deployment1. Entry looper rail alignment is set by fixing the distance between the rails Implemented during (QFD) in the analyze phase. to 2,200 mm, with a tolerance less than 0.5mm. June 2003 shutdown. QFD in the analyze phase.2. Aligning roll one, which was skewed by 2 mm, sets hot bridle roll Implemented during Response surface method alignment. Roll two was found to be working fine. June 2003 shutdown. in the improve phase.3. Pressure in the snubber roll has a direct impact based on the thickness. Implemented during Regarding the contour of the speed factor, the correct pressure in the September 2003 shutdown. snubber rolls is set to achieve the optimal speed factor. I I 43S I X S I G M A F O R U M M A G A Z I N E NOVEMBER 2004

Project Leads to $1.65 Million in Annual Savings In the end, the vital few enablers were the thickness with 73 data points. So, the vital few X’s are length,of the coil, length of the coil, weight of the coil and thickness, weight and pressure difference in the snub-pressure difference in the snubber rolls. (Snubber ber roll.rolls are two rolls that maintain the pressure differ-ence.) To finalize the identity of the vital few X’s (rea- We did a design of experiments (DOE) taking threesons for side tracking), hypothesis testing was done. of the earlier mentioned factors to two levels with no replicates. We didn’t consider the length of the coil The testing was based on a single Y and multiple X’s. for DOE because it comes from another mill, so thisBecause the X’s were discrete, we performed a T-test, mill had no control over that X factor.with the results shown in the Pareto chart in Figure 11(p. 42). Here the p-value is assumed as 0.05 (95% con- We drew a response surface (see Figure 12, p. 43)fidence level). The test data was taken for one week, based on the results of the DOE between the speed factor, thickness of the coil and pressure difference inFigure 14A. Speed factor at CGL1 the snubber rolls. As you can see, to achieve an opti- For July 2003 mum speed factor as the thickness increases, the pres- sure difference between the snubber roll should 1.5 Moving range Individual value Upper control decrease. However, the response surface 3-D diagram limit = 1.200 showed the pressure difference should not go above 1.0 4.0 because a drastic fall of the speed factor will cause Mean = 0.9807 the coil to fail, resulting in side tracking. 0.5Subgroup 0 Lower control The pilot solutions to the side tracking problem are limit = 0.7618 shown in the Figure 13 (p. 43). After implementing 100 200 300 400 500 600 700 the recommendations, the sigma level improved to 2.24. Production of galvanized coils increased by 620 0.8 Upper control tons a month. The average margin on each galvanized 0.7 coil became more than $2,200. 0.6 0.5 limit = 0.2690 The project therefore resulted in an annual recur- rent audited savings of about $1.65 million. Control 0.4 R = 0.08232 charts showing results after implementing a few of the 0.3 recommendations are shown in the Figure 14A and 0.2 14B. Figure 15 is a before and after cost benefit com- 0.1 Lower control parison. 0.0 limit = 0 ACKNOWLEDGMENTS CGL1 = Continuous galvanizing line one The author thanks the following for assisting with this project and article:Figure 14B. Speed factor at CGL1 J.C. Johnston, regional director, Global Services Center, P&O Nedlloyd; For August 2003 Bridget Irvine, senior account manager, Business Services Division, P&O Nedlloyd; Chris McGeogh, global process manager, shipment management, Moving range Individual value 1.0 Upper control Global Services Center, P&O Nedlloyd; H.C. Kharkar, chief, cold rolled limit = 1.167 mill, Tata Steel; Avneesh Gupta, Master Black Belt, flat products, Tata Steel; 0.5 and Mohit Ratolikar, process manager, CGL1, Tata Steel. Mean = 0.9429 0.0 NOTES Subgroup 0 Lower control limit = 0.7186 100 200 300 1.0 Upper control limit = 0.2756 0.5 R = 0.08435 Lower control 0.0 limit = 0Figure 15. Before and After Comparison 1. A description of the Anderson Darling test can be found at www.itl.nist. gov/div898/handbook/prc/section2/prc213.htm.Parameters Before the project After the project 2. A description of the Kolmogorov-Smirnov test can be found at www.itl. (December 2002 (July 2003 to nist.gov/div898/handbook/eda/section3/eda35g.htm.Defects per million to March 2003) August 2003)oppor tunities WHAT DO YOU THINK OF THIS ARTICLE? Please shareSigma level (short-term) 759,774 235,525 your comments and thoughts with the editor by e-mailingAverage speed factor 1.50 2.24 [email protected] of speed factor 0.874 0.969 24.02% 76.44%44 I IN O V E M B E R 2 0 0 4 WWW.ASQ.ORG