The process of stamping stainless steel housing components for large household appliances had a very high level of waste. Numerous optimization measures were taken, but the average reject rate remained around 20%, and the variation between subsequent production shifts was very high. While nearly flawless production runs were observed, there were also those with defects as high as 30 percent. This led to constant attempts to answer the question of why a given production shift was so good or so bad. This, in turn, led to the implementation of various measures that not only failed to change the result but actually increased the observed variation. The results of the scrap rate [%] for 50 consecutive production shifts are shown in the ImR Chart below.
Chart 1
The mR chart tells us that subsequent production changes differ by an average of 8.19%, but differences of 26.77% should not surprise us. Any difference between two subsequent changes that does not exceed this value should not be interpreted as an improvement or deterioration of the process. Therefore, this change is not due to a single cause that could be addressed with appropriate actions, but to the system – how the process is currently set up.
Chart I, on the other hand, tells us that on average we deliver 19.85% of waste. However, we should not be surprised by results ranging from a perfect 0% to an extremely poor 41.64% – although their probability is low. This is still common cause variation and should not be interpreted as a signal of improvement or deterioration. Nearly 70% of future results should be expected to fall within the range of 12.59 to 27.11% waste, and 95% of future results will fall within the range of 5.33 to 34.47% bad components. So we have a stable predictable process with an unacceptable mean and huge variation.
Stamping process – short description
The process begins with applying foil to the steel to protect its surface in subsequent stages. The steel is delivered in coils, and the foil is also supplied. The foil has the same width as the steel strip. After applying the foil, the steel is cut crosswise into steel forms and stored on pallets, which will serve as the raw material for the actual stamping operation. This is schematically illustrated in Figure 1.
Figure 1
Figure 2 shows the further steps of this process:
Figure 2
- The robot picks up the steel form from the pallet and places it on a station where oil is applied to its surface.
- Oil is applied. The oil’s role is to reduce friction, reduce material flow resistance, and stabilize plastic deformation.
- After passing through two transport stations, the steel form is placed on a mold, where shaping will take place. First, the tool closes and it is held in the area of the so-called drawbead, which are visible on the plan view of the form as yellow lines. Then, the actual stamping operation takes place, where the pre-held form is deformed at a specified power and speed.
- Next, subsequent operations begin – punching holes that are functionally required or will be used for assembly of the finished product.
- The final punching die cuts off the excess portion of the form, which was used to hold it in place during the previous operations.
- The conveyor belt delivers the next produced parts to the operator, who removes the protective foil and then visually assesses each piece and decides whether it continues or is scrapped.
Project goal: reduction of current waste by at least 50%
The first question to ask was why components were rejected after the stamping process. The operator assessed each one – 100% inspection – for the presence of:
- Foil dents: these are defects that occur during operation 6 (see Figure 2) – the actual stamping operation. The foil, especially around the drawbead, would tear. Small pieces would fall off, and if they got caught between the steel form and the stamping tool, a dent would form after the operation.
- Chip dents: the same mechanism, but caused by steel scraps that were created during subsequent stamping operations. The solution for this is known. An increase in the amount of chips means the tools need to be sharpened, and the problem disappears.
- Material defects: steel quality
- Abrasions and scratches: visual defects most often caused by improperly applied protective foil
Additionally, once an hour, the operator checked three pieces for the presence of micro-cracks.
The defects included in the scrap rate per production shift shown in Chart 1 are 96% foil dents, hence this problem needed to be addressed.
What factors influence the formation and number of foil dents?
As usual, there were many theories. Here are a few:
- Quality results vary depending on the steel supplier. There were two steel suppliers. Operators claimed that the process performed better or worse depending on the steel supplier. The problem was that the preferences for the supplier varied from operator to operator.
- We should have used a thicker protective foil with a stronger adhesive. A thicker foil and a stronger adhesive would have reduced the frequency of small pieces of foil peeling off, which caused dents. The problem was that the foil thickness and adhesive strength had already been increased, which some claim helped, but operators reported having to use considerable force to remove the foil after the process and couldn’t imagine increasing it further.
- The stamping power and speed should be lower. The more “gentle” we handle the form, the less chance there is of a piece of foil peeling off.
- The most peeling foil is found in the area of the drawbead. We don’t need to protect this area with foil, as it’s cut off after die-cutting anyway. It may be scratched or have scuff marks. Let’s apply a narrower foil to eliminate the amount of scrap that could potentially cause dents (Figure 3).
Figure 3
5. The biggest problem is that a single piece of foil can cause a series of defective pieces. We should increase the frequency of preventive cleaning of the stamping tool. This should help reduce the number of dents caused by the same piece of foil. This solution was implemented as soon as it was first conceived. It didn’t change the overall situation, only reduced production capacity – but since it had many supporters, it became a part of daily routine.
6. We will increase the oil viscosity, which should result in the foil scrap being left on the form (Figure 4) rather than on the stamping tool (Figure 5). A single foil scrap that wanders between the form and the stamping tool can cause one defective piece, or even 13 if it sticks to the stamping tool and the operator sees it, stops the line, and cleans the tool. Of course, there are also intermediate situations
Figure 4
Figure 5
These were strong theories and extensive engineering knowledge. For some reason, despite this knowledge, the problem remained unsolved. In such situation, it becomes clear that the root causes lie not only in individual causes, but also that we need to understand their interaction. The only way I know of to do this is to conduct an experiment – DoE.
We planned DoE
The goal of DoE will be to understand how the theories mentioned above affect the number of foil dents. Within a single DoE, we want to create runs ranging from very high dent counts to those where, ideally, there are none at all.
The factors and levels of testing are presented in the Prediction Table below.
Choosing an Experimental Design. To test six factors at two levels each, we have four scenarios to choose from:
- Full-factorial design 26=64
- Fractional design 26-1VI=32
- Fractional design 26-2IV=16
- Fractional design 26-3III=8
Scenarios 1 and 2 require too many runs. We reject Scenario 4 due to Resolution III, which is characterized by a very high level of confounding. This leaves us with Scenario 3, which requires 16 runs and whose Confounding Structure is consistent with Resolution IV. Resolution IV is characterized by main effects being confounded with third-order interactions, and second-order interactions being confounded with each other. While inferring the main effects should not pose any problems, we should pause here and consider what second-order interactions are possible. These predictions will be useful in analyzing this DoE. As a reminder, a second-order interaction occurs when the effect of a given factor on Y varies depending on the setting of the second factor.
The predicted second-order interactions are:
- AF, oil viscosity * lubrication frequency
- BC, stamping speed * stamping power
- BD and CD, steel supplier * stamping parameters (power and speed)
- DE, steel supplier * foil width
The Confounding Structure for the 26-2IV=64 experiment is given below:
A + BCE + DEF + ABCDF
B + ACE + CDF + ABDEF
C + ABE + BDF + ACDEF
D + AEF + BCF + ABCDE
E + ABC + ADF + BCDEF
F + ADE + BCD + ABCEF
AB + CE + ACDF + BDEF
AC + BE + ABDF + CDEF
AD + EF + ABCF + BCDE
AE + BC + DF + ABCDEF
AF + DE + ABCD + BCEF
BD + CF + ABEF + ACDE
BF + CD + ABDE + ACEF
ABD + ACF + BEF + CDE
ABF + ACD + BDE + CEF
What will we be measuring? In a process like stamping, it’s difficult to assess the impact of changing parameters on just a few pieces. Therefore, and given the time window available for conducting this DoE, we assumed we would produce at least 50 pieces in each run. If there are many dents, we’ll stop; if we’re doing well, we’ll produce 200 pieces. Each piece will be assessed on an ongoing basis and either be scrapped or further processed.
The Ys from this DoE will be:
- Y1: total scrap rate [%] (all defect types combined)
- Y2: foil scrap rate [%]
- Y3: number of bad parts caused by a single foil scrap (see Figures 4 and 5).
The limitations of conclusions are presented in the Factor Relationship Diagram below. As a reminder, the those limitations indicate the conditions under which the experimental conclusions will be valid in practice.
The experiment will be conducted on a single stamping line/machine. The protective foil will be from a single supplier, reflecting the current situation. Other parameters not tested in this DoE will be maintained at levels consistent with current technology. All parts will be evaluated by a single operator who successfully passed the Measurement System Evaluation for visual assessment of parts after stamping. Because changing the oil viscosity requires cleaning the machine and waiting for all the previously used oil to “extract” from the system, we will only change this factor once. For the same reasons, we will use one barrel of each oil.
For the experiment, we will use steel and foil from different deliveries, representing the variation of the input material we encounter daily in the process.
We conducted the DoE
After creating the experimental matrix according to the 26-2IV=16 design, we proceeded to conduct the experiment. The experiment recipe is ready – all that’s needed is to follow the plan and collect previously planned observations. It’s also important to conduct the experiment in random order whenever possible and record the actual order of runs.
In our case, we have a randomization constraint. First, all runs were performed for the low-viscosity oil (we will randomly conduct 8 runs for this factor A level). Then, the system will be purged and we will conduct the remaining 8 runs (also randomly) for the higher-viscosity oil. The DoE matrix along with the results for individual Ys is shown in the table below
Practical DoE Analysis
Did we create variation in Y1 (total scrap rate) and Y2 (foil scrap rate)?
Chart 2
Chart 2 shows Y1: total scrap rate [%] and Y2: foil scrap rate [%] for each run from minimum to maximum. Both Ys created variation that significantly exceeds the common cause variation. This means that the tested factors influence both total scrap rate and the foil scrap rate. The analysis we will perform shortly will tell us how to reduce the phenomenon of foil stripping, which causes dents.
Y1 and Y2 are very strongly correlated (see Chart 3). This means that the tested factors did not increase other types of waste, and as is the case in everyday life, foil dents are the main and dominant defect. Analyzing both Ys is pointless, as the same factors will affect each in the same way, so we will only analyze Y1.
Chart 3
Did we create variation in Y3 (number of bad parts caused by a single foil scrap)?
Chart 4
Variation created in this DoE ranges from 1 to an average of 10 bad pieces due to a single piece of foil (see Chart 4). This is a very high variation, meaning the tested factors influence whether a piece of foil remains on the form (only one dent is created) or a series of dents are formed. Analyzing this DoE will tell us not only how to limit the number of fallen off pieces (Y1 analysis) but also how to ensure that any foil remains on the form.
Y1 Analysis: Which factors and interactions cause foil dents?
Chart 5
Effect on total scrap rate [%]
B: stamping power [kN] 9,21
C: stamping speed [mm/s] 12,74
E: foil width -17,69
B: stamping power [kN]*C: stamping speed [mm/s] 9,61
C: stamping speed [mm/s]*D: steel supplier -11,11
R-sq R-sq(adj)
73,83% 60,74%
The effect size Y1 indicates the percentage by which we can influence the scrap rate. Therefore, we will designate the Main Effects B, C, and E as active, as well as the two interactions BC and CD. These five effects can explain almost 74% of the variation created in this doe.
Chart 6
Effect E: Changing the width of the protective foil to a narrower one causes the average waste to decrease from 25.31% to 7.62%. This factor is not involved in any active interaction, meaning it will behave similarly regardless of the settings of the other factors tested in this doe.
Effects B and C are involved in an active interaction – therefore, their proper interpretation is only possible after analyzing this interaction.
Chart 7
BC interaction (Chart 7): When the stamping speed is 30 mm/s, the stamping power is insignificant and the scrap rate percentage is approximately 10%. However, when the stamping speed is 40 mm/s, then increasing the stamping power from 220 to 350 kN causes the scrap rate percentage to increase from 13.43% to 32.35%.
Chart 8
CD Interaction (Chart 8): When we stamp steel from supplier A, reducing the stamping speed from 40 to 30 mm/s reduces the average scrap rate from 29.3% to 5.45%. In the case of supplier B, this effect does not occur and the average scrap rate remains at approximately 15%.
Chart 9
Considering the above main effects and interactions, to reduce the scrap rate to approximately 5%, a narrower foil should be used. The absence of foil within the drawbeads significantly reduces the detachment of foil fragments, which cause dents. Furthermore, reducing the stamping speed to 30 mm/s allows for greater tolerance to differences between steel suppliers, and the stamping power can remain at any of the tested levels.
Analysis Y3: What factors and interactions cause foil scraps to remain on the form? How can we avoid producing a series of bad parts caused by a single foil?
Chart 10
Effect on number of bad parts caused by 1 scrap of foil [piece]
A: oil viscosity[mm2/s] -2,463
B: stamping power [kN] 2,613
D: steel supplier 2,937
R-sq R-sq(adj)
75,32% 69,15%
We consider the D, B and A effects as active. With these three effects we can explain 75% of the variation created in this doe in Y3.
Chart 11
Effect A: Increasing the oil viscosity from 80 to 200 mm²/s reduces the number of bad parts per foil scrap from an average of 5.5 to an average of 3 bad parts.
Effect B: Reducing the stamping power from 230 to 220 kN reduces the number of bad parts per foil scrap from an average of 5.5 to an average of 2.9 bad parts.
Effect D: Steel supplier A produces an average of 2.7 bad parts per foil scrap, and supplier B produces an average of 5.7 bad parts per foil scrap.
Chart 12
By increasing the oil viscosity to 200 mm²/s and reducing the stamping power, we eliminate differences between steel suppliers. The number of bad parts caused by a single piece of foil will then average around 1.5 pieces.
Summary and recommended settings taking into account the conclusions from the Y1 and Y3 analysis:
To reduce the frequency of foil scraps that cause dents, use a narrower foil to avoid it being in the drawbeads. Additionally, to avoid differences between steel suppliers, use a lower stamping speed.
To reduce the number of bad parts caused by a single detached foil scrap (so that it doesn’t remain on the stamping tool but is released with the form), use a higher-viscosity oil. Furthermore, to avoid differences between steel suppliers, use a lower stamping power.
Recommended settings:
- Oil viscosity 200 mm²/s
- Stamping speed 30 mm/s
- Stamping power 220 kN
The new settings were implemented immediately after the DoE analysis. Chart 13 shows a comparison of the results before (50 production shifts) and after (another 50 shifts). Average waste decreased from 19.85% to 2.63%. Shift-to-shift variation also decreased significantly!
Chart 13
ARTICLE IN PDF
Have questions, thoughts, or something that particularly piques your interest? We invite you to join the discussion below on our LinkedIn post. Follow us to stay up to date.
. If you’d like to learn how to design and analyze experiments yourself, check our Six Sigma Green Belt, Black Belt, and DOE (basic and advanced) training courses.
