This document describes experiments to optimize pipetting parameters for a robotic liquid handler. A screening design of experiments (DOE) was used to evaluate factors affecting pipetting accuracy and precision. Conditioning and aspiration volume had the largest effects on bias and coefficient of variation (%CV). A second DOE with fewer factors confirmed these results and found interactions between aspiration delay, conditioning, and dispense speed. A confirmation run showed settings predicted from the model effects achieved much tighter %CV than original settings or best settings from the DOE runs. While bias could not be improved, the experiments demonstrated pipetting precision can be optimized through experimental design.
Optimization of Pipetting Parameters for a Robotic Liquid Handler
1. Optimization of Pipetting Parameters for
a Robotic Liquid Handler
Kristi Ballard, Chuck Kemmerer, Thorsten Verch
14 April 2013
Temple QA/RA DOE Course
2. Problem Statement
• Pipetting steps have associated accuracy/bias and
precision ranges
• Minimizing bias and %CV is key to analytical method
performance
• Factors affecting pipet bias and precision of a robotic
liquid handler were investigated for improvement
opportunities
• Goal: Lowest Bias and Lowest %CV possible
3. Accuracy/Bias & Precision
Image:
http://academics.wellesley.edu
/Chemistry/Chem105manual/A
ppendices/uncertainty_analysi
s.html
Created By: Adilia James '07
and Sarah Coutlee '07
• Bias can be accommodated through calibration
• Imprecision cannot be adjusted
4. TECAN Robotic Liquid Handler
8 Independent syringes to
aspirate/dispense
LiHa picks up tips and moves liquid
from reservoirs into microwell plates
Dispense into microwell plates:
Screening: 4 runs / plate
Confirmation: 2 runs / plate
5. Artel Measurement System
• Measurement of a red
dye in an undiluted blue
background
• Calculates delivered
volume by comparing
measured concentration
with expected
concentration
• Lambert-Beer:
Image:
J Biomol Screen July 12, 2012 , doi: 10.1177/1087057112453433
c: Concentration
A: Absorbance (measured)
e: Extinction Coefficient (known)
d: Path length (known)
6. Selected Pipetting Parameters
Aspirate Speed ↑
Dispense Speed ↓
Blow-out (Leading Airgap)
Volume
Aspirate Delay ↑
Dispense Delay ↓
Break-off Speed
Held Constant:
System Air Gap
Trailing Air Gap
Calibration Factors
Pipet Height in Liquid
Channels
Tip Size
Liquid Type / Viscosity
(Volume)
Retract Speed post Aspirate
Retract Speed post Dispense
Pre-wet (Conditioning)
7. Screening DOE 1
Objectives:
Comparison of different instruments
Evaluation of 11 factors
8. Experimental Design (Half)
RunOr
der
Asp
vol
Asp
spd
Asp
del
• 12+12 Run Plackett-Burman fold-over
11 Factors
• 1 Block for robot
• Second half of the fold-over design
was run on a separate day
Asp ret
spd
Asp ld
air
gap
Trl air
gap
Cond
vol
Dis
spd Dis dly
Dis ret
spd
Dis bk-off
spd
1 200 50 50 5 10 10 yes 50 1000 60 20
2 200 150 50 60 10 0 yes 50 50 5 200
3 25 50 50 5 0 0 No 50 50 5 20
4 200 150 1000 5 10 10 No 500 50 5 20
5 25 150 50 5 0 10 yes 500 50 60 200
6 25 150 1000 60 0 10 yes 50 1000 5 20
7 25 50 50 60 10 10 No 500 1000 5 200
8 25 50 1000 60 10 0 yes 500 50 60 20
9 200 150 50 60 0 0 No 500 1000 60 20
10 200 50 1000 5 0 0 yes 500 1000 5 200
11 200 50 1000 60 0 10 No 50 50 60 200
12 25 150 1000 5 10 0 No 50 1000 60 200
• 8 channel replicates (pipets) / run
3 dispense replicates / channel
4 robots
12 runs x 3 replicates x 8 wells x 4
robots = 1152 data points!
9. Results
DOE1 DOE2 Boxplot of Log %Bias
BBR0479 BBR0508 BBR0509 EVO150
50
25
0
-25
-50
-75
Robot ID
% Volume Bias
Boxplot of % Volume Bias
• Difference in the mean bias between the robots
depends on run
• Subset of instruments was used as representative
• Higher variability observed in one instrument
• Blocking by instrument was not used downstream in
order to maintain a “worst case” scenario
Data were averaged across instruments
BBR0508 BBR0509
BBR0508 BBR0509
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
Robot ID
Log %CV
Boxplot of Log %CV
1.75
1.50
1.25
1.00
0.75
0.50
Robot ID
Log %Bias
10. Data Transformation
Residual Plots for Absolute % volume bias
Normal Probability Plot Versus Fits
-20 0 20 40 60
Histogram Versus Order
900
800
700
600
500
400
300
200
100
• Log transformed data distribution is closer to normal
• Log data used for all models
99.99
99
90
50
10
1
0.01
Residual
Percent
7 8 9 10
60
45
30
15
0
Fitted Value
Residual
-10 0 10 20 30 40 50
240
180
120
60
0
Residual
Frequency
1100
1000
1
60
45
30
15
0
Observation Order
Residual
11. Results, Main Factor Model, Pareto
Log % Bias Log % CV
DOE1 with Volume, Pareto, Main Factors Only
(response is Log %Bias, Alpha = 0.05)
Conditioning
Target Volume (uL)
Retract Speed
Disp Speed
Disp Delay
Dispense Trailing Air Gap
Asp lead/Disp Trail Air Gap
Break-off Speed
Asp Retract Speed
Asp delay
Asp Speed
2.030
0 1 2 3 4
DOE1 with Volume, Pareto, Main Factors Only
(response is Log %CV, Alpha = 0.05)
Target Volume (uL)
Asp delay
Conditioning
Asp lead/Disp Trail Air Gap
Break-off Speed
Asp Retract Speed
Dispense Trailing Air Gap
Asp Speed
Disp Speed
Retract Speed
• Aspiration Volume has the largest effect for %CV
2.030
→ Not practical to improve process by volume. Required to pipette
multiple volumes. *Use the small volume for future runs*
• Conditioning has the largest effect for %Bias
→ Do not condition during future runs.
Term
Standardized Effect
Disp Delay
0 1 2 3 4 5 6 7
Term
Standardized Effect
12. Results, Effects
• Positive conditioning
mean high bias and high
%CV
→ Do not condition
during future runs.
• Blocked for volume
• Main effects only
13. Screening DOE 2
Objectives:
Confirmation of DOE1
Improved Experimental Design
Remove Run Block
14. Experimental Design (Half)
• 12+12 Run Plackett-Burman fold-over
9 Factors
• Low volume only (20 mL)
• 8 channel replicates (pipets) / run
3 dispense replicates / channel
2 robots
24 runs x 3 replicates x 8 wells x 2
robots = 1152 data points!
Run Asp Spd Asp del
Asp Retr
Spd
Asp ld/Dsp
Trl Air Gap Cond Disp Spd Disp Del Ret Spd
BkOff
Spd
1 50 1000 60 10 no 50 50 60 20
2 250 1000 60 10 no 500 1000 5 20
3 250 50 5 10 no 50 1000 60 200
4 50 50 60 10 yes 50 1000 5 200
5 50 50 60 0 yes 500 1000 60 20
6 250 1000 60 0 yes 50 50 5 200
7 250 50 60 0 no 500 50 60 200
8 50 1000 5 10 yes 500 50 60 200
9 50 1000 5 0 no 500 1000 5 200
10 50 50 5 0 no 50 50 5 20
11 250 1000 5 0 yes 50 1000 60 20
12 250 50 5 10 yes 500 50 5 20
15. Results, Main Effects Model, Pareto
Log % Bias Log % CV
Pareto Chart of the Standardized Effects
(response is Log %CV, Alpha = 0.05)
Conditioning
Disp Speed
Asp lead/Disp Trail Air Gap
Break-off Speed
Asp delay
Retract Speed
Asp Speed
Disp Delay
Asp Retract Speed
2.024
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Term
Standardized Effect
Conditioning
Disp Speed
Asp Speed
Asp lead/Disp Trail Air Gap
Retract Speed
Asp Retract Speed
Break-off Speed
Asp delay
Disp Delay
0.0 0.5 1.0 1.5 2.0
• Conditioning & Dispense Speed have the largest effect for both responses
Term
Standardized Effect
2.024
Pareto Chart of the Standardized Effects
(response is Log %Bias, Alpha = 0.05)
16. Results, Second Order Model, Pareto
Log % Bias Log % CV
E
AG
F
D
BE
AF
AJ
BC
BF
AH
AE
J
B
AC
AD
H
A
AB
G
E
F
AG
BE
AJ
AF
AD
BC
A
D
AH
H
C
J
BF
AE
B
G
F actor Name
• Conditioning has the largest effect for both responses
• Many interactions
→ Follow up with a resolution V design
C
2.052
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Term
Standardized Effect
F actor Name
A A sp Speed
B A sp delay
C A sp Retract Speed
D A sp lead/Disp Trail A ir Gap
E C onditioning
F Disp Speed
G Disp Delay
H Retract Speed
J Break-off Speed
DOE2, Pareto
(response is Log %CV, Alpha = 0.05)
AC
AB
0.0 0.5 1.0 1.5 2.0 2.5
Term
Standardized Effect
2.052
A A sp Speed
B A sp delay
C A sp Retract Speed
D A sp lead/Disp Trail A ir Gap
E C onditioning
F Disp Speed
G Disp Delay
H Retract Speed
J Break-off Speed
DOE2, Pareto
(response is Log %Bias, Alpha = 0.05)
17. Main Effects
Main Effects Plot for Log %CV
50 250
Main Effects Plot for Log %Bias
Data Means
50 1000 5 60
no y es 50 500
5 60 20 200
0.9
0.8
0.7
Data Means
50 1000 5 60
0 10
0.9
0.8
0.7
no y es 50 500
50 1000
0.9
0.8
0.7
5 60 20 200
A sp Speed
Mean
A sp delay A sp Retract Speed
A sp lead/Disp Trail A ir Gap C onditioning Disp Speed
Disp Delay Retract Speed Break-off Speed
A sp Speed
50 250
1.1
1.0
0.9
0 10
1.1
1.0
0.9
50 1000
1.1
1.0
0.9
Largest Effects
(Minimize Bias and %CV)
• Trailing Airgap (high)
• Conditioning (low)
• Dispense Speed (high)
Mean
A sp delay A sp Retract Speed
A sp lead/Disp Trail A ir Gap C onditioning Disp Speed
Disp Delay Retract Speed Break-off Speed
18. Interaction
Effects
Asp delay
Conditioning
50 1000 no yes 50 1000 20 200
1.0
0.8
01..06
0.8
01..06
0.8
01..06
0.8
0.6
Asp Speed
Asp delay
Conditioning
Disp Delay
Break-off Speed
Asp
Speed
50
250
Asp delay
50
1000
Conditioning
no
yes
Disp
Delay
50
1000
Interaction Plot for Log %CV
Data Means
50 1000 no yes 50 1000 20 200
Largest Effects
(Minimize Bias and %CV)
• Trailing Airgap (high)
• Conditioning (low)
• Dispense Speed (high)
1.10
0.95
0.80
1.10
0.95
0.80
1.10
0.95
0.80
1.10
0.95
0.80
Asp Speed
Asp delay
Conditioning
Disp Delay
Break-off Speed
Asp
Speed
50
250
50
1000
no
yes
Disp
Delay
50
1000
Interaction Plot for Log %Bias
Data Means
Asp
Speed
Asp
Delay
Conditioning
Disp
Delay
Break-off
Speed
Asp
Speed
Asp
Delay
Conditioning
Disp
Delay
Break-off
Speed
21. Set-Up
• 3 Liquid Classes:
1. Best conditions from DOE runs
2. Best conditions predicted by effects model
3. Original settings
• 2 calibration modifications:
1. Original
2. Reduced off-set
• 2 Volumes:
1. 20 mL
2. 100 mL
• 6 dispense replicates / channel
• 8 channels
• 6 * 8 = 24 data points per liquid class & volume
22. Results (Total Averaged Data)
% Bias % CV
• No improvement in %Bias
• Much tighter %CV for settings predicted by the model
• Single outlier in original liquid class
No solution was added to the wells (robotic failure?)
23. Results (Outlier Excluded)
% Bias % CV
DOE Data DOE Data-Offset DOE Prediction DOE Prediction-Offset Old
10.0
7.5
5.0
2.5
0.0
-2.5
-5.0
Liquid Class
Average % Bias
Boxplot of Average % Bias, outlier excluded
DOE Data DOE Data-Offset DOE Prediction DOE Prediction-Offset Old
700
600
500
400
300
200
100
0
-100
-200
Liquid Class
Average %CV of bias
Boxplot of Average %CV of bias, outlier excluded
• No improvement in %Bias
• Much tighter %CV for settings predicted by the model
24. Conclusions
- Strategy -
• No significant difference between robots (?)
Liquid classes can be transferred
• Optimization at 20 mL can be transferred to larger volumes
• Bias may be addressed with calibration
Liquid class can be optimized for precision
• Placket-Burman design is efficient to screen a number of factors
and reduce the total. Fold-over design helps understanding main
factors
• Even with a limited model, improvements can be achieved
• Numerous interactions require a fractional factorial design for
complete modeling
25. Conclusions
- DOE Analysis -
• Response may need log-transformation
• Bias and Precision are driven by different factors with some
overlapp
• Most factors interact. Few stand-alone main effects.
• Confounded models may not adequately predict
performance
• Blocking for factors that are not going to be optimized
further can enhance analysis of the remaining factors
26. Path Forward
• Characterization DOE and eventually a Response Surface
DOE with reduced number of factors
• Repeat the study for a protein solution (2-5% BSA)
• Improve experimental design based on experience with this
study:
– No conditioning
– Focus on lowest volume
– Consider a fractional factorial design with less replication for screening
28. DOE1 –
99
90
50
DOE1, Residual Plots for Log %Bias
Normal Probability Plot Versus Fits
-0.50 -0.25 0.00 0.25 0.50 Residual Plots
10
1
Residual
Percent
0.4 0.6 0.8 1.0 1.2
0.50
0.25
0.00
-0.25
-0.50
Fitted Value
Residual
Histogram Versus Order
-0.4 -0.2 0.0 0.2 0.4
24
18
12
6
0
Residual
Frequency
1 5 10 15 20 25 30 35 40 45
0.50
0.25
0.00
-0.25
-0.50
Observation Order
Residual
29. DOE2 –
DOE2, Residual Plots for Log %CV
Normal Probability Plot Versus Fits
-0.50 -0.25 0.00 0.25 0.50 Residual Plots
99
90
50
10
1
Residual
Percent
0.3 0.6 0.9 1.2
0.50
0.25
0.00
-0.25
-0.50
Fitted Value
Residual
Histogram Versus Order
-0.6 -0.4 -0.2 0.0 0.2 0.4
8
6
4
2
0
Residual
Frequency
1 5 10 15 20 25 30 35 40 45
0.50
0.25
0.00
-0.25
-0.50
Observation Order
Residual
DOE2, Residual Plots for Log %Bias
Normal Probability Plot Versus Fits
-0.50 -0.25 0.00 0.25 0.50
99
90
50
10
1
Residual
Percent
0.6 0.8 1.0 1.2 1.4
0.50
0.25
0.00
-0.25
-0.50
Fitted Value
Residual
Histogram Versus Order
-0.4 -0.2 0.0 0.2 0.4
12
9
6
3
0
Residual
Frequency
1 5 10 15 20 25 30 35 40 45
0.50
0.25
0.00
-0.25
-0.50
Observation Order
Residual
30. Comparison
of DOE1 & 2,
Main Effects
DOE1, Main Effects Plot for Log %Bias
50 150
1.0
0.9
0.8
Data Means
50 1000 5 60 0 10
0 10
1.0
0.9
0.8
no y es 50 500 50 1000
5 60
1.0
0.9
0.8
20 200
A sp Speed
Mean
A sp delay A sp Retract Speed A sp lead/Disp Trail A ir Gap
Dispense Trailing A ir Gap C onditioning Disp Speed Disp Delay
Retract Speed Break-off Speed
DOE2, Main Effects Plot for Log %Bias
50 250
1.1
1.0
0.9
Data Means
50 1000 5 60
0 10
1.1
1.0
0.9
no y es 50 500
50 1000
1.1
1.0
0.9
5 60 20 200
A sp Speed
Mean
A sp delay A sp Retract Speed
A sp lead/Disp Trail A ir Gap C onditioning Disp Speed
Disp Delay Retract Speed Break-off Speed
31. Comparison
of DOE1 & 2,
Main Effects
DOE1, Main Effects Plot for Log %CV
50 150
0.5
0.4
0.3
Data Means
50 1000 5 60 0 10
0 10
0.5
0.4
0.3
no y es 50 500 50 1000
5 60
0.5
0.4
0.3
20 200
A sp Speed
Mean
A sp delay A sp Retract Speed A sp lead/Disp Trail A ir Gap
Dispense Trailing A ir Gap C onditioning Disp Speed Disp Delay
Retract Speed Break-off Speed
DOE2, Main Effects Plot for Log %CV
50 250
0.9
0.8
0.7
Data Means
50 1000 5 60
0 10
0.9
0.8
0.7
no y es 50 500
50 1000
0.9
0.8
0.7
5 60 20 200
A sp Speed
Mean
A sp delay A sp Retract Speed
A sp lead/Disp Trail A ir Gap C onditioning Disp Speed
Disp Delay Retract Speed Break-off Speed
32. DOE1, Results, Second Order Factor Model
2.064
F actor Name
A Target V olume (uL)
B A sp Speed
C A sp delay
D A sp Retract Speed
E A sp lead/Disp Trail A ir Gap
F Dispense Trailing A ir Gap
G C onditioning
H Disp Speed
J Disp Delay
A
C
G
AD
AG
E
L
D
F
B
H
AF
AK
AJ
BC
AC
AL
K
AE
J
AH
2.064
• Aspiration Volume has the largest effect for %CV
F actor Name
A Target V olume (uL)
B A sp Speed
C A sp delay
D A sp Retract Speed
E A sp lead/Disp Trail A ir Gap
F Dispense Trailing A ir Gap
G C onditioning
H Disp Speed
J Disp Delay
→ Not practical to improve process by volume. Required to pipette
multiple volumes. *Use the small volume for future runs*
• Conditioning has the largest effect for %Bias
→ Do not condition during future runs.
G
A
AH
AE
AK
AF
AG
BC
K
AD
AJ
H
J
AL
F
E
L
D
AB
C
B
AC
0 1 2 3 4
Term
Standardized Effect
K Retract Speed
L Break-off Speed
DOE1 with Volume, Pareto
(response is Log %Bias, Alpha = 0.05)
AB
0 1 2 3 4 5 6 7
Term
Standardized Effect
K Retract Speed
L Break-off Speed
DOE1 with Volume, Pareto
(response is Log %CV, Alpha = 0.05)
33. DOE2 – ANOVA of log %CV
Analysis of Variance for Log %CV (coded units)
Source DF Seq SS Adj SS Adj MS F P
Main Effects 9 2.15259 2.15259 0.23918 1.73 0.116
Asp Speed 1 0.00360 0.00360 0.00360 0.03 0.873
Asp delay 1 0.02056 0.02056 0.02056 0.15 0.702
Asp Retract Speed 1 0.00006 0.00006 0.00006 0.00 0.983
Asp lead/Disp Trail Air Gap 1 0.46353 0.46353 0.46353 3.35 0.075
Conditioning 1 1.01453 1.01453 1.01453 7.34 0.010
Disp Speed 1 0.61547 0.61547 0.61547 4.45 0.042
Disp Delay 1 0.00050 0.00050 0.00050 0.00 0.952
Retract Speed 1 0.00440 0.00440 0.00440 0.03 0.859
Break-off Speed 1 0.02993 0.02993 0.02993 0.22 0.644
Residual Error 38 5.25335 5.25335 0.13825
Lack of Fit 14 2.57768 2.57768 0.18412 1.65 0.136
Pure Error 24 2.67567 2.67567 0.11149
Total 47 7.40594
Unusual Observations for Log %CV
Obs StdOrder Log %CV Fit SE Fit Residual St Resid
40 40 1.33029 0.59572 0.16971 0.73456 2.22R
R denotes an observation with a large standardized residual.
Estimated Coefficients for Log %CV using data in uncoded units
Term Coef
Constant 1.03053
Asp Speed 0.000086644
Asp delay -4.35743E-05
Asp Retract Speed -0.00004191
Asp lead/Disp Trail Air Gap -0.0196538
Conditioning 0.145383
Disp Speed -5.03271E-04
Disp Delay -6.78313E-06
Retract Speed -0.00034799
Break-off Speed 0.000277455
Factorial Fit: Log %CV versus Asp Speed, Asp delay, ...
Estimated Effects and Coefficients for Log %CV (coded units)
Term Effect Coef SE Coef T P
Constant 0.7983 0.05367 14.87 0.000
Asp Speed 0.0173 0.0087 0.05367 0.16 0.873
Asp delay -0.0414 -0.0207 0.05367 -0.39 0.702
Asp Retract Speed -0.0023 -0.0012 0.05367 -0.02 0.983
Asp lead/Disp Trail Air Gap -0.1965 -0.0983 0.05367 -1.83 0.075
Conditioning 0.2908 0.1454 0.05367 2.71 0.010
Disp Speed -0.2265 -0.1132 0.05367 -2.11 0.042
Disp Delay -0.0064 -0.0032 0.05367 -0.06 0.952
Retract Speed -0.0191 -0.0096 0.05367 -0.18 0.859
Break-off Speed 0.0499 0.0250 0.05367 0.47 0.644
S = 0.371815 PRESS = 8.38208
R-Sq = 29.07% R-Sq(pred) = 0.00% R-Sq(adj) = 12.27%
34. DOE2 – ANOVA of log %Bias
Analysis of Variance for Log %Bias (coded units)
Source DF Seq SS Adj SS Adj MS F P
Main Effects 9 1.23678 1.23678 0.137420 1.22 0.312
Asp Speed 1 0.08264 0.08264 0.082638 0.73 0.397
Asp delay 1 0.00415 0.00415 0.004153 0.04 0.849
Asp Retract Speed 1 0.02543 0.02543 0.025431 0.23 0.637
Asp lead/Disp Trail Air Gap 1 0.06964 0.06964 0.069644 0.62 0.436
Conditioning 1 0.49750 0.49750 0.497495 4.42 0.042
Disp Speed 1 0.49657 0.49657 0.496573 4.41 0.042
Disp Delay 1 0.00262 0.00262 0.002625 0.02 0.879
Retract Speed 1 0.04862 0.04862 0.048619 0.43 0.515
Break-off Speed 1 0.00961 0.00961 0.009605 0.09 0.772
Residual Error 38 4.27884 4.27884 0.112601
Lack of Fit 14 1.78955 1.78955 0.127825 1.23 0.316
Pure Error 24 2.48929 2.48929 0.103721
Total 47 5.51562
Unusual Observations for Log %Bias
Obs StdOrder Log %Bias Fit SE Fit Residual St Resid
40 40 1.53174 0.85781 0.15316 0.67393 2.26R
44 44 1.74424 1.07701 0.15316 0.66723 2.23R
48 48 1.72926 1.01078 0.15316 0.71848 2.41R
R denotes an observation with a large standardized residual.
Estimated Coefficients for Log %Bias using data in uncoded units
Term Coef
Constant 1.03024
Asp Speed 0.000414925
Asp delay -1.95826E-05
Asp Retract Speed 0.00083700
Asp lead/Disp Trail Air Gap -0.00761819
Conditioning 0.101806
Disp Speed -4.52052E-04
Disp Delay 0.000015568
Retract Speed 0.00115731
Break-off Speed -1.57179E-04
Factorial Fit: Log %Bias versus Asp Speed, Asp delay, ...
Estimated Effects and Coefficients for Log %Bias (coded units)
Term Effect Coef SE Coef T P
Constant 0.9755 0.04843 20.14 0.000
Asp Speed 0.0830 0.0415 0.04843 0.86 0.397
Asp delay -0.0186 -0.0093 0.04843 -0.19 0.849
Asp Retract Speed 0.0460 0.0230 0.04843 0.48 0.637
Asp lead/Disp Trail Air Gap -0.0762 -0.0381 0.04843 -0.79 0.436
Conditioning 0.2036 0.1018 0.04843 2.10 0.042
Disp Speed -0.2034 -0.1017 0.04843 -2.10 0.042
Disp Delay 0.0148 0.0074 0.04843 0.15 0.879
Retract Speed 0.0637 0.0318 0.04843 0.66 0.515
Break-off Speed -0.0283 -0.0141 0.04843 -0.29 0.772
S = 0.335561 PRESS = 6.82718
R-Sq = 22.42% R-Sq(pred) = 0.00% R-Sq(adj) = 4.05%
35. % Bias Interaction Plots: DOE1 vs. DOE 2
DOE1 DOE2
50 1000 no yes 50 1000 20 200
1.10
0.95
0.80
1.10
0.95
0.80
1.10
0.95
0.80
1.10
0.95
0.80
Asp Speed
Asp delay
Conditioning
Disp Delay
Break-off Speed
Asp
Speed
50
250
Asp delay
50
1000
Conditioning
no
yes
Disp
Delay
50
1000
Interaction Plot for Log %Bias
Data Means
50 1000 No yes 50 1000 20 200
6
0
-6
6
0
-6
6
0
-6
6
0
-6
aspiration speed*
aspiration delay
conditioning volume**
dispense delay
Dispense Break-off Speed***
aspiration
speed*
50
150
aspiration
delay
50
1000
conditioning
volume**
No
yes
dispense
delay
50
1000
Interaction Plot for Average Run Bias (%)
Data Means
• 2nd Level interactions are confounded in the Plackett Burman Design
• Results from both DOE studies indicate interactions are significant
• A screening study with clear 2-level interactions is needed to learn more
• Mostly similar effects. Differences may be due to:
DOE 1 run with volume as a factor
DOE 2 had a reduced number of factors (9) from DOE (11)
36. % Bias Interaction Plots: DOE1 vs. DOE 2
DOE1 DOE2
50 1000 No yes 50 1000 20 200
102
96
90
102
96
90
102
96
90
102
96
90
aspiration
speed*
aspiration
delay
conditioning
volume**
dispense
delay
• 2nd Level interactions are confounded in the Plackett Burman Design
• Results from both DOE studies indicate interactions are significant
• A screening study with clear 2-level interactions is needed to learn more
• Mostly similar effects. Differences may be due to:
DOE 1 run with volume as a factor
DOE 2 had a reduced number of factors (9) from DOE (11)
aspiration speed*
aspiration delay
conditioning volume**
dispense delay
Dispense Break-off Speed***
50
150
50
1000
No
yes
50
1000
Interaction Plot for Average Run Accuracy (%)
Data Means
50 1000 no yes 50 1000 20 200
1.0
0.8
01..06
0.8
01..06
0.8
01..06
0.8
0.6
Asp Speed
Asp delay
Conditioning
Disp Delay
Break-off Speed
Asp
Speed
50
250
Asp delay
50
1000
Conditioning
no
yes
Disp
Delay
50
1000
Interaction Plot for Log %CV
Data Means