Improved Applesauce Yields through Design of Experiments (DOE) Karl Hofman Director – Rapid Continuous Improvement Dr Pepper Snapple Group
Improved Applesauce Yields through
Design of Experiments (DOE)
Karl Hofman
Director – Rapid Continuous Improvement
Dr Pepper Snapple Group
3 Takeaways
2
Premium Tea
Gourmet CSDs
Juice & Juice DrinksFlavored CSDs
DPS is the Market Leader in many Categories
Mixers
3
Rapid Continuous Improvement (RCI)
A Culture Change Journey
4
We are creating a culture of breakthrough change by:
•Setting breakthrough goals (GOAL deployment)
•Measuring progress and addressing root cause (CAP)
•Developing Lean leaders (tracks)
5
5,300 participants -- 500 Kaizens – 100 locations
5
RCI Results
Balanced Approach to RCI:
• Safety – 37% reduction in recordables
• Quality – Consumer complaints down 25%
• Delivery – Fill Rate 99.9%
• Productivity – Reduced warehouse
footprint by 2.5MM ft2 (25%)
• Growth – +50 bps in WD SSMP share
Sustaining our World class status in Asset Utilization
58%
lower
than
peers
Improving Applesauce Yield
Using DOE and Minitab to Drive Improvement
Williamson, NY – Home of Motts’ Applesauce
7
Multiserve Applesauce
(MSAS) Line
Singleserve Applesauce
(SSAS) Lines (3)Pouch Lines (2)
Applesauce Process Flow Overview
8
KPIV's (X's) KPOV's (Y's)
Ratio Ratio of Bertocchi
Local Rate (throughput) / Chopped Fruit
Speed of Auger
Rotor Speed Bertocchi apple stream
Auger Speed Apple waste
Rotopulse Speed
Local Rate (throughput)
Speed
Screen Size Chopped Apples
Local Rate (throughput)
Apple Size
Speed Bertocchi/
Condition of Stator & Chopped Apples
Temperature
Local Rate (throughput) Heated Sauce
Temperature Heated Sauce
Local Rate (throughput)
Pressure
Cooked Sauce
Cook Time Cooked Sauce
Clearance of Screens Cooked Sauce
Clearance of Paddles
Position of Paddles
Screensize
Cooked Sauce
Waste
Applesauce Process Flow
Start
Apple Hoppers
Chopper
Bertocchi
End
PC Pump
Steam Pick 1
Steam Pick 2
Back Pressure Valve
Cook Tank
Finisher
Cooked Sauce heading to Pot
Finisher Waste Stream
Cooked Sauce to the Lines
DOE #1
DOE #2
Bertocchi Extractor – Description
The apple extraction processes for all applesauce lines in
Williamson incorporate the use of a specific brand of extractor
known as a Bertocchi extractor. Pictures of the Bertocchi used for
the MSAS line and SSAS lines are shown below with key
components identified:
9
Bertocchi Unit
(opened with rotor and screen removed)
Bertocchi Rotor and
Screen
(removed and separated)
Rotor
Screen
Yield Loss/Waste Measurement
A manual yield loss measurement process was developed enabling
the team to directly measure the waste at each Bertocchi unit (MSAS
and SSAS):
10
1. Bertocchi waste
chute modification
allows capture of
waste stream
2. Timed waste
collection
3. Weight
measured and
yield loss
calculated based
on output rate
Manual waste measurement process enables analysis of Bertocchi parameters to optimize
performance and reduce losses. Same method is applied to the finisher process.
Question? Is this an acceptable measurement system? Need to perform an MSA….
Yield Loss/Waste Measurement
• Initial Gage R&R study
conducted with 3 operators as
a nested study and using 15
second measurement intervals
Gage R&R Study - Nested ANOVA
Gage R&R (Nested) for Result
Source DF SS MS F P
Operator 2 7.7205 3.8602 0.34504 0.733
Batch (Operator) 3 33.5631 11.1877 5.28396 0.040
Repeatability 6 12.7038 2.1173
Total 11 53.9873
Gage R&R
%Contribution
Source VarComp (of VarComp)
Total Gage R&R 2.11729 31.83
Repeatability 2.11729 31.83
Reproducibility 0.00000 0.00
Part-To-Part 4.53520 68.17
Total Variation 6.65249 100.00
Study Var %Study Var
Source StdDev (SD) (6 × SD) (%SV)
Total Gage R&R 1.45509 8.7305 56.42
Repeatability 1.45509 8.7305 56.42
Reproducibility 0.00000 0.0000 0.00
Part-To-Part 2.12960 12.7776 82.57
Total Variation 2.57924 15.4755 100.00
Number of Distinct Categories = 2
Yield Loss/Waste Measurement
• Repeat Gage R&R study
conducted with 2 operators as
a nested study and using 30
second measurement intervalsGage R&R (Nested) for Results
Source DF SS MS F P
Operator 1 0.018 0.0176 0.000 0.987
Sample (Operator) 2 108.542 54.2709 111.455 0.000
Repeatability 12 5.843 0.4869
Total 15 114.403
Gage R&R
%Contribution
Source VarComp (of VarComp)
Total Gage R&R 0.4869 3.49
Repeatability 0.4869 3.49
Reproducibility 0.0000 0.00
Part-To-Part 13.4460 96.51
Total Variation 13.9329 100.00
Study Var %Study Var
Source StdDev (SD) (6 × SD) (%SV)
Total Gage R&R 0.69780 4.1868 18.69
Repeatability 0.69780 4.1868 18.69
Reproducibility 0.00000 0.0000 0.00
Part-To-Part 3.66688 22.0013 98.24
Total Variation 3.73268 22.3961 100.00
Number of Distinct Categories = 7
Bertocchi Extraction – Screen Life
13
• Current practice for Bertocchi process is to replace mechanical
screen at 1 month (SSAS) and 2 month (MSAS) intervals.
• Comparison of older (end of life) screen and new (beginning of life)
screen showed significant waste differences (simple comparative
experiment on MSAS Bertocchi unit of old vs. new screen):
Screen life impacts yield
significantly. Comparison of old
vs. new screen on the MSAS
Bertocchi showed a ~60% improvement.
Old Screen MSAS
New Screen MSAS
0.020.010.00-0.01-0.02-0.03-0.04-0.05
MS
AS
Scre
en
Yield Loss (Normalized)
Boxplot of Yield Loss (Normalized) vs MSAS Screen
Old Screen MSAS
New Screen MSAS
0.0300.0250.0200.01 50.01 0
P-Value 0.678
P-Value 0.506
Multiple Comparisons
Levene’s Test
MS
AS
Scre
en
Test for Equal Variances: Yield Loss (Normalized) vs MSAS ScreenMultiple comparison intervals for the standard deviation, α = 0.05
If intervals do not overlap, the corresponding stdevs are significantly different.
Mood Median Test: Yield Loss (Normalized) versus MSAS Screen
Mood median test for Yield Loss (Normalized)
Chi-Square = 6.56 DF = 1 P = 0.010
Individual 95.0% CIs
MSAS Screen N≤ N> Median Q3-Q1 +---------+---------+---------+------
New Screen MSAS 5 0 -0.0293 0.0237 (---------*--------)
Old Screen MSAS 3 7 0.0051 0.0131 (-----*--)
+---------+---------+---------+------
-0.045 -0.030 -0.015 0.000
Overall median = 0.0000
* NOTE * Levels with < 6 observations have confidence < 95.0%
A 95.0% CI for median(New Screen MSAS) - median(Old Screen MSAS): (-0.0523,-0.0200)
Bertocchi Extraction – DOE #1
14
• Evaluation of equipment components showed screen life impact on yield – what about
process set-up and operating conditions?
• DOE (Design of Experiments) allows us to evaluate multiple process inputs, determine
which inputs are critical and then reduce waste by optimizing those critical inputs in a
minimal number of experimental runs.
• We identified 4 key inputs to evaluate (Bertocchi feed rate, Rotor Speed, Infeed Auger
Speed and Rotopulse Speed). The optimal DOE design was determined to be a 4 factor, 1/2
fractional factorial design with 2 center points and 2 replicates for a total of 18 runs – See
Table Below:
This DOE was
performed in
March 2015 on the
MSAS Bertocchi
unit with a new
screen using the
manual method of
waste
measurement
Bertocchi Extraction – DOE #1
15
DOE results showed that Rotor Speed was the main factor affecting waste with Bertocchi
Feed Rate having a minor impact on waste compared to Rotor Speed. All other factors
had insignificant impact on waste.• Rotor Speed (B) is the major
factor affecting waste
• Bertocchi feed rate (A) is a minor
factor affecting waste
• Other factors (C&D) and
interactions do not impact waste
• Increasing Rotor Speed
significantly reduces waste
• Decreasing Bertocchi Feed Rate
slightly reduces waste
Evaluating the DOE #1 Model for Bertocchi Yield
• Evaluation of power of the experiment and residuals
16
The power of the experiment is 74% and
reasonable given the nature of our
product (it’s applesauce). The presence
of heteroscedasticity in the residuals
was identified but not concerning after
replicating optimal settings post DOE –
Bottom line was that the DOE yielded a
positive result that was demonstrated
and sustained in process post DOE.
Power and Sample Size
2-Level Factorial Design
α = 0.05 Assumed standard deviation = 0.34
Factors: 4 Base Design: 4, 8
Blocks: none
Including a term for center points in model.
Center Total
Points Effect Reps Runs Power
2 0.5 2 18 0.744813
Putting DOE #1 Results into Practice
• Prior to the DOE, Bertocchi extractor screens were replaced based on time and
multiple inputs on the Bertocchi extractor were adjusted to reduce losses with
unpredictable and inconsistent results.
• Post DOE, it was established that screen wear and rotor speed for the Bertocchi
extraction process were the only critical inputs affecting yield. Control charts were
established on the Bertocchi waste stream to monitor yield at a set frequency and
adjust rotor speed as needed to maintain process control until screen replacement
was necessary.
17
An I-MR Chart is used to track
yield loss at the Bertocchi
extraction process based on
waste stream measurement. As
yield loss increases, the
Bertocchi rotor speed is
adjusted until reaching it’s
practical process limit then the
screen is changed. Note that as
the process has continued, the
UCL/LCL have tightened,
showing a very consistent
process yield.
DOE #2 – Chopping and Finishing Process
• The second DOE was conducted on the chopping/finishing stream that is
mixed with the Bertocchi stream to make our MSAS and SSAS products.
The key factors considered were Bertocchi ratio (ratio of Bertocchi stream
to chopped/finished stream) temperature (for chopped/finished stream)
and tank level (for cooking tank for chopped/finished stream).
• Based on the factors and perceived significant interactions, a 2 level 3
factor full factorial design was selected with 2 center points and 2
replicates for a total of 18 runs for DOE #2.
18
This DOE was
performed in
July 2016 on
the MSAS line
using a new
screen for the
finisher
process
Run # StdOrder RunOrder CenterPt Blocks Ratio (Bertocchi) Cook Temp (Pick 1) Cook Tank Level
1 17 4 0 1 80 220 2
2 13 1 1 1 70 210 3
3 9 2 1 1 70 210 0
4 8 3 1 1 90 230 3
5 10 5 1 1 90 210 0
6 5 6 1 1 70 210 3
7 2 7 1 1 90 210 0
8 11 8 1 1 70 230 0
9 15 9 1 1 70 230 3
10 12 10 1 1 90 230 0
11 6 11 1 1 90 210 3
12 1 12 1 1 70 210 0
13 7 13 1 1 70 230 3
14 4 14 1 1 90 230 0
15 14 16 1 1 90 210 3
16 16 17 1 1 90 230 3
17 3 18 1 1 70 230 0
18 18 15 0 1 80 220 2
DOE #2 Results
• DOE #2 showed that the only significant factor affecting overall process yield was
the ratio of Bertocchi stream to chopped/finished stream. The higher the ratio, the
lower the yield losses.
• Evaluation of the model showed no issues with residuals and a power of 74% which
was acceptable (again, it’s applesauce)
19
Power and Sample Size
2-Level Factorial Design
α = 0.05 Assumed standard deviation = 0.017
Factors: 3 Base Design: 3, 8
Blocks: none
Including a term for center points in model.
Center Total
Points Effect Reps Runs Power
2 0.025 2 18 0.744813
Next Steps
• Increase the Bertocchi to Chopped/Finished
Ratio to further optimize yield
•Shore up measurement system (perform further
MSA studies to identify where we can reduce
gage error)
• Implement similar process controls as
Bertocchi (I-MR charting/SPC) on yield losses at
finishing process step
•Analyze incoming apple crop to understand key
inputs and identify potential opportunities to
further improve yields
20
3 Takeaways
21
Questions?
22