Improving Properties of Steel Using Basic Tools of Quality

1
Improving Properties of Steel Using Basic Tools
of Quality

• Case Study in Manufacturing

2
Overview

• Project aimed to eliminate internal rejections of
  a product used for exposed parts for several
  major manufacturers.
• Case study describes the use of a assortment of
  quality improvement tools and strategies used to
  bring this about.
• Also describes how the inevitable issues that
  occur while running a complicated industrial
  experiment were handled.

3
Shifting a Distribution
4
Problem Description Limitations

• Yield strength (YS) of the steel sheets exceeded
  specified maximum of 255 MPa 28 of the time.
  Customer occasionally experienced splitting on
  formed parts.
• Ability to lower YS constrained by minimum
  tensile strength (TS) requirement and tensile
  strength after a post-stamping step.
• High cost of failures prohibited generating
  off-spec in trials.
• Measurement system issues for some key process
  variables

5
Trade-off Between Yield Tensile Strength
6
Process Map
Molten Steel
Slab Casting
Secondary Refining
Hot Rolling
Chem
Chem

• Temp1 (N)
• Temp2 (C)
• Temp3 (C)
• Cooling Practice (C)

• Ingredient A (C)
• Ingredient B (C)
• Ingredient C (C)
• Ingredient D (N)
• Ingredient E (N)

• Turndown
• A (N)

• Ingred. A
• Pickup (N)

Pickle/ Cold Roll
Coating
Temper Rolling
Storage
Mech. Props

• Anneal T (C)
• Cooling Section Ts (C)

• Elongation (F)

• Age Hardening (N)

7
Laws of Chemistry Physics Say

• Increasing Ingredient A increases YS TS.
• Increasing Ingredient B increases YS TS.
• Increasing Ingredient C increases YS TS.
• Impact of D and E depends upon level of the other
  three. Generally, more of either increases YS
  TS.
• Higher hot rolling temps (especially T2 T3)
  give lower YS TS.
• Increment of A has bigger effect than one of B,
  which is bigger than an increment of C.

8
Laws of Management Say

• Cant change recipe outside current ranges
  without customer approval.
• Multiple customers. Cant make changes that
  require multiple recipes.
• Limited reapplication potential. Cant make
  stuff that no one can use.
• Heats of steel cost 10s of 1000s each. Make
  each one count.
• Arent you done yet?? (Dont take forever to fix
  the product).

9
Restrictions Imply

• Normal 2k factorials poor choiceslikely to make
  lots of unusable steel. (Big sigma means
  alarmingly bold settings or large n).
• EVOP a possibility, if reasonably sure of the
  factors included. Limited to 2 or 3, though, due
  to length of time needed at each corner of the
  design matrix.
• Large standard deviation implies long trial
  duration. What if we picked the wrong variables?

10
Possible Courses of Action

• Change one factor at a time (OFAT). Rejected
  because it would take too much time, also because
  sub-optimal solutions are typical outcomes, even
  in best-cases.
• Change all suspected factors simultaneously to
  better levels (All-FAT). Gives feel of
  decisiveness, but what has been learned if
  problem is solved?
• Run factorial design with least risky candidate
  variables. Include riskier variables in
  subsequent iterations if necessary.

11
Selected Factorial Approach

• Engineering knowledge allowed team to pick low
  risk variables at meaningfully different levels
  in 22 factorial HM Cooling Practice A B
  normal and reduced Ingredient C addition.
• Had data for one corner already and could begin
  generating for the new cooling practice almost
  immediately. New heats made on orders for
  customer with less restrictive requirementsno
  risk of making coils with no homes.

12
Phase I Results
255 (13.5) 82
250 (10.6) 250

• Conclude
• All combos fail.
• Lo C better??
• U worse??

High
Ingredient C Level
248 (11.3) 248
245 (15.0) 138
Low
Flat
U-Profile
Cooling Practice
13
Phase II Plan

• Stop adding Ingredient A to ladle. (Addition made
  to assure TS after final part processing was
  still OK.
• Keep reduced level of C, since it made the
  reducing A less risky (heats also slightly
  cheaper to make, too).
• Add more Ingredient B to minimize risk of failing
  tensile strength standard. (Chose two higher
  levels).
• Use U-profile, since some still believed it was
  better option.

14
Phase II Design Layout
Phase I
248 (11.3) 248
Ingredient A Added
Phase II
No Ingredient A Addition
Highest B
Higher B
Base B
15
Phase II Design YS Results

• Conclude
• No A addition
• meets spec.
• B adds strength.
• Omit A addition,
• use base B if TS
• is okay.

Phase I
248 (11.3) 248
Ingredient A Added
Phase II
242 (7.4) 14
235 (8.5) 31
244 (14.4) 17
No A Addition
Base B
Highest B
Higher B
16
Phase II Design TS Results

• Conclude
• Base B will often
• fail TS Spec
• Middle B will rarely
• fail TS Spec.
• Use Middle B level
• in longer term trials.

Phase I
355 (7.7) 248
A Added
Phase II
353 (6.1) 14
346 (5.2) 31
358 (6.4) 17
No A Addition
Base B
Highest B
Higher B
17
YS Distributions, Base vs New Recipe
New C Lower B Higher Ladle A no
Old C High B Base Ladle A yes
255 MPa max.
255 MPa max.
28 gt 255 MPa
5 gt 255 MPa
N 330
N 105
18
TS Distributions, Base vs New Recipe
Old C High B Base Ladle A yes
New C Lower B Higher Ladle A no
340 MPa min.
340 MPa min.
N 330
N 105
19
Final Steps to Attain YS Max

• Steelmakers changed practices allowing upper
  limit for Ingredient A to be reduced by 25.
  Effectively capped maximum YS near where we
  wanted it. Still had 5 internal rejections,
  though.
• Measurement error for Ingredient A was known to
  be high. Values above specific limit now
  automatically rechecked. Heats that exceed limit
  on second test are applied to orders for
  customers where the 255 MPa does not apply.
• Internal failures for YS rarely have rarely
  occurred in the past year since these last
  changes were made.

20
Reducing Variation
21
New Tensile Strength Issue

• Recall that YS has strong positive correlation to
  TS. Small percentage (3-5) miss minimum
  requirement since project was completed.
• Low values occur tend to occur in sporadic
  clusters at bottom of cycles noted on trend
  plots.
• Commonly believed to occur because of cyclical
  pattern in values of Ingredient A, but
  steelmakers cant do more than they already are,
  and anything done to increase TS mean will raise
  YS mean.

22
What to Do?

• Look at other variables on the process map. Can
  any of them be changed to move TS mean and not
  the YS mean?
• If the mean cant be moved, then variance has to
  be reduced.

23
Components of Variance

• Method attempts to quantify the amount of the
  total observed variation associated with the
  components included in the study.
• Focuses efforts on understanding how to reduce
  the biggest component first.
• Moves on to next biggest if goal hasnt been
  achieved, or if cost associated with reducing
  biggest term is prohibitive or cant be done
  immediately.

24
Components of Variance Design
1
2
3
n
Heat
HM LU
1
2
1
2
1
2
1
2
I/N LUs
1
2
1
2
1
2
1
2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
Lab MSE
25
COV Results

• 77 of variance in heat-to-heat component.
• 16 of variance occurs between hot mill lineups.
• 7 of variance due to MSE.
• Note impact of Heat 3 on magnitude of
  Heat-to-Heat.

26
Are the Results Valid?

• Reality checkdoes strength level for heat make
  sense?
• If it does, then focus on root causes of variance
  in the Steelmaking area.
• If not, investigate further. In this case, we
  found that material processed in unplanned manner
  downstream. Low TS values processed on same LU,
  and are the only ones that did. Steelmaking and
  Coating are confounded.
• Lesson Use common sense. Seeing neednt always
  mean believing.

27
Evaluation of Historical Data

• This type of data potentially has value,
  especially if one is looking for trends in
  processes or outputs.
• Not generally a substitute for designed
  experiments, though data mining software makes it
  possible to extract more information than was
  possible in the past.
• Main interest here initially was in determining
  ranges of potential key variables. Large ranges
  could help identify variables likely to inflate
  variance.

28
Ingredient A Variation

• Found that hot mill temperature variation was
  small consistent with COV results no smoking
  gun at the hot mill.
• The experiments from the previous year had
  identified Ingredient A as a key factor in YS
  TS. Small change had significant impact on
  observed values. Range noted in historical data
  was wide.
• Extracted sets comprised of heats with the
  highest and lowest 10 of Ingredient A values.
  31 of coils from lowest A heats failed. 3 from
  highest A heats failed.

29
TS by Ingredient A Classification
Lowest A
Middle A
Highest A
30
TS Cycles from High/Low Heats Generally Track
Each Other
Tensile Strength Trend by Ingredient A Level
Highest A
Lowest A
31
Next Steps

• Controlling for Ingredient A level over time
  suggests downstream origin for variance. It
  wont be possible to eliminate heats with lowest
  values of Ingredient A.
• Task becomes identifying which coating line
  variables have ranges most likely to impact TS,
  quantifying the magnitudes of the effects, and
  developing plans to reduce the impacts.

32
Conclusion

• Case study has presented application of two
  problem solving methods.
• In the first, designed experiments were used to
  shift a process average to nearly completely
  eliminate internal rejections. Modification of
  application practices completed the task.
• The second application used tools aimed at
  reducing variation when shifting the mean was not
  an option.
• Which set one uses first depends upon where
  process is relative to where it needs to be.

33
Questions?