DOE (Design of Experiments) for Powder Coating Thickness

T

Toefuzz - 2006

#1
Greetings! I recently completed my first gage r and r and am now getting ready to conduct my first D.O.E.

My goal is to determine which variables have the biggest impact on coating thickness in our powder coat shop. The five variables I want to test are:

Gun to Part Distance
Powder Output
Transport Air
Booth Air
Gun Voltage

My plan is to coat one rack of parts for each run (as directed by my software) and take the mean coating thickness of that rack after measuring each of its 42 parts. The objective is to find which factors most influence coating thickness and to also find an optimal setting to achieve a desired millage.

A customer recommended I try using the Design Ease software so I have that downloaded and am all set to go, except I am not sure which type of D.O.E. to use. I've read quite a few articles this evening and am leaning towards the 2 Level Factor Design experiment, but have heard others say the Taguchi OA method is better. In addition to these two methods I have the following methods available:

Irregular Fraction General Factorial D-Optimal Plackett Burman

I guess my questions for you are - Am I going about this correctly? Do you think this study will meet my objectives? Which study should I use?

I am relatively new to statistics and know next to nothing about D.O.E. so any advice you might be able to offer would be greatly appreciated.
 
Elsmar Forum Sponsor

Tim Folkerts

Super Moderator
#2
A few quick thoughts.

A 2 level, full factorial DOE would require 32 runs for 5 factors and would be the simplest design conceptually.

A 2 level fractional factorial could cut that to 16 runs and you would still get info on two-factor interactions. (For example, perhaps high booth air is good, but only when the transport air is also high.)

If you just want main effects, then you need a minimum of six run. A fractional factorial DOE with 8 trials would work well here.

The taguchi designs L32, L16, and L8 should be equivalent to the factorial designs above. So you can call them taguchi designs for the customer who suggests taguchi designs, and call then fractional factorials desings for anyone else who perfers that terminology. :rolleyes:


Try to pick at least on run to repeat to see how consistently the system works at one setting.


Besides the mean thickness, you might study the effects of the factors on the standard deviation of the thickness, or even on the minimum thickness. You presumably don't just want a particular average thickness, but you also want uniform thickness.

Bottom line - I would suggest some sort of two level factorial design. Whether you do 8, 16, or 32 runs depends on how much info you really want, and how expensive it is to do the experiments.


By the way. Welcome to the Cove. I hope I haven't overloaded you yet!

Tim
 
T

Toefuzz - 2006

#3
Thanks for the prompt response!

Seeing as how this is my first DOE I'm rather eager to get things going so I think I will go with the 2 Level Factorial Design using 16 experiments. I hadn't really thought of inputing multiple responses (effects) but normally track the mean, std. deviation, minimum, and maximum so adding those to the study would probably be very worthwhile... It will be interesting to see what the final reports and analysis look like with so many different responses. All of the tutorials I've seen thus far seem to only use one response.

Again, thanks for the advice... Glad to hear I'm on the right path!
 
T

Toefuzz - 2006

#4
Okie doke, here's another question for ya'll.

One of my variables is gun to part distance... This is simply how far away from the parts my guns are positioned. This distance dictates how the recipricator (simply moves guns up and down over a given distance at a given pace) is setup. The spray pattern exiting a gun is funnel shaped, so at 20 inches from the parts the pattern might be 20 inches wide, but if the guns are 8 inches from the part the pattern might only be 6 inches wide (numbers are just a made up example). B/c the spray pattern is narrowed the reciprocator settings need to be changed to ensure an even spray pattern over the rack.

My question for you is, how do I go about testing the gun to part distance? Can I change the recipricator settings when I move the guns in or does this invalidate my study?

Thanks!
 

Miner

Forum Moderator
Staff member
Admin
#5
Toefuzz said:
One of my variables is gun to part distance... This is simply how far away from the parts my guns are positioned. This distance dictates how the recipricator (simply moves guns up and down over a given distance at a given pace) is setup. The spray pattern exiting a gun is funnel shaped, so at 20 inches from the parts the pattern might be 20 inches wide, but if the guns are 8 inches from the part the pattern might only be 6 inches wide (numbers are just a made up example). B/c the spray pattern is narrowed the reciprocator settings need to be changed to ensure an even spray pattern over the rack.

My question for you is, how do I go about testing the gun to part distance? Can I change the recipricator settings when I move the guns in or does this invalidate my study?
You can take one of two approaches. One approach would be to treat recipricator settings as another factor. If you do not have a good understanding of how to adjust one to compensate for the other, this would give you good information on how to adjust recipricator settings in relation to the gun distance.

A second approach is to link the two and treat them as a single variable. If you do understand how to adjust the recipricator settings for a gun distance, this might be the way to go. If you don't, you really won't understand the relationship any better when you are done, and changes you make in the recipricator setting may offset or overpower changes in the gun distance.
 
T

Toefuzz - 2006

#6
For now we've decided to ignore the gun to part distance and will treat it as a constant (takes quite a bit of work to change it). In the future I think it will be very interesting to perform a seperate DOE using the variables related to gun to part distance to see which have the biggest impact.

I've been discussing this experiment with a knowledgable customer and a few questions have been raised that he won't be able to answer before I plan on beginning the experiment tomorrow morning.

My plan is to conduct a 2 level factorial design with 4 variables. A full factorial with no replications results in 16 experiments. Adding a replication results in 32 experiments, which would require a little more time than I had planned on spending. A half fractorial results in 8 experiments with a replication taking things up to 16 experiments.

Now from what I understand the full fractorial (with so few variables) will do a better job of estimating the main effects and two factor interactions. Adding the replication will allow me to compute the estimates of pure error.

My question for you is this... If my goal is to determine which of my variables has the greatest affect on millage and I only want to run 16 experiments am I better off conducting the full factorial or the half factorial with replication? Also, would it truly be worthwhile to take the time to conduct the 32 experiments necessary to conduct a full factorial with replication?

Thanks again for all your help!
 

Tim Folkerts

Super Moderator
#7
I think the choice comes down to a judgement call. What do you think has more effect on the outcome - interactions between the factors or other uncontrolled variations (setting the factors, time of day, equipment getting warmed up or worn out)?

With 8 runs and replications, you will get minimal information about interactions. You may find that some interaction is important, but not know which one. With 16 runs and no replications, you will know about the interactions, but uncontrolled variations might confuse the results and hide the true causes of variation.

Sorry I can't provide a definite answer. You may just have to go with your gut.

Also, if you have time and the equipment is adjustable, you might try running the center point a few times. This will provide some info about replication and also confirm the results.


Tim F
 
T

Toefuzz - 2006

#8
Tim Folkerts said:
I think the choice comes down to a judgement call. What do you think has more effect on the outcome - interactions between the factors or other uncontrolled variations (setting the factors, time of day, equipment getting warmed up or worn out)?

With 8 runs and replications, you will get minimal information about interactions. You may find that some interaction is important, but not know which one. With 16 runs and no replications, you will know about the interactions, but uncontrolled variations might confuse the results and hide the true causes of variation.

Sorry I can't provide a definite answer. You may just have to go with your gut.

Also, if you have time and the equipment is adjustable, you might try running the center point a few times. This will provide some info about replication and also confirm the results.


Tim F
Things make more sense after reading your post Tim. I'm expecting to find several interactions that are important with uncontrolled variations being minimal, especially uncontrolled variations of the effects I'm interested in. There will be some important 'noise' (humidity, temp, etc) but it's nothing I can control and really nothing I can even test at this time of year.

I do like your idea of running the center point a few times... I think I'll go with the 16 experiments, no replication and two or three center points (which should result in 18 or 19 experiments total).

Should be an interesting day tomorrow!
 

Miner

Forum Moderator
Staff member
Admin
#9
Consider the following noise factors that may impact your longer term results:
  • Lot-to-lot changes in powder (e.g., moisture content)
  • Periodic drops in air pressure when demands is greater than supply
  • Day to day changes in humidity impacting an electrostatic application

I would recommend replicating the experiment over several days to include these factors even at the price of a larger experiment. Since interactions are expected, make sure the experiment is Resolution IV at a minimum, and be aware that 2-factor interactions can confound with other 2-factor interactions.
 
Q

qualeety

#10
before you proceed....

one thing you want to do before you proceed with your first DOE is....

implement spc on powder weight to ensure you have a stable system.

As a enamel engineer, i noticed great deal of differences between "powder weights" when all other parameters are same (or assumed to be same)

Unless you are assured of the stable powder delivery system, your DOE results may not mean much.

Futhermore, you might want to ensure the humidity is constant (as you mentioned)...i noticed drastic difference bewteen summer and winter....we had to install environmentally controlled room to ensure all powders are climatized before they are used.

one more thing...have you discussed the powder size with your suppliers?...if not, you should....we made more improvement with consistent powder size than any other changes we implmented.

here is another trick...if you are getting black spots on your finished products, think of installing "magnets" on the top of conveyor...we found "metal dust" from the conveyor chain falls on the product and magnets help remove them.

good luck with your doe and hope your experiment would be more beneficial than mine was....
 
Thread starter Similar threads Forum Replies Date
J DoE (Design of Experiments) - Multiple responses with different factors Using Minitab Software 2
K Power point training slides on DOE (Design of Experiments) Wanted Training - Internal, External, Online and Distance Learning 2
N DOE (design of experiments) in Minitab 16 Using Minitab Software 6
P When can we apply DOE (Design of Experiments)? Six Sigma 8
P Cyclic Effect Factoring in DOE (Design of Experiments) Quality Tools, Improvement and Analysis 1
J How to Estimate the Number of Replicates on a DOE (Design of Experiments) Six Sigma 5
K Minitab Lack-of-Fit Test in DOE (Design of Experiments) Using Minitab Software 4
M Taguchi L4 Orthogonal Array Design of Experiments (DOE) in an Excel spreadsheet Excel .xls Spreadsheet Templates and Tools 3
V Linking the FMEA Causes & Action Plan with DOE Experiments or Experimentation Design FMEA and Control Plans 1
B Taguchi DOE (Design of Experiments) example using Minitab Using Minitab Software 3
C 3 Factor DOE (Design of Experiments) Template Document Control Systems, Procedures, Forms and Templates 0
L DOE (Design of Experiments) suggestions Using Minitab Software 7
L DOE (Design of Experiments) - Analyse Variation with Minitab Using Minitab Software 5
E 5^2 DOE (Design of Experiments) with 3 Replications, 5 Levels 2 Factors Statistical Analysis Tools, Techniques and SPC 6
M DOE (Design of Experiments) for 3 Variables, Many Levels Quality Tools, Improvement and Analysis 2
J DOE (Design of Experiments) Dependent Factors Using Minitab Software 1
C DOE (Design of Experiments) with 'Uncontrollable' Factors Using Minitab Software 5
Q DOE (Design of Experiments) for Accounting - New to Minitab need help Using Minitab Software 5
M DOE (Design of Experiments) - CQE Primer question Professional Certifications and Degrees 4
C Training courses for DOE (Design of Experiments) Training - Internal, External, Online and Distance Learning 3
P Taguchi method using Minitab - DOE (Design of Experiments) Help Quality Tools, Improvement and Analysis 7
S Sample Size Attribute DOE (Design of Experiments) Six Sigma 9
M Excel files for DOE (Design of Experiments) Excel .xls Spreadsheet Templates and Tools 6
T Minitab - Mixture DOE (Design of Experiments) - Sandability of a Gelcoat Using Minitab Software 5
R Custom DOE (Design of Experiments) Analysis in JMP Quality Tools, Improvement and Analysis 11
P Need help on DOE (Design of Experiments) - EPS molding process Quality Assurance and Compliance Software Tools and Solutions 2
A Good books on DOE (Design of Experiments) - Recommendations wanted Book, Video, Blog and Web Site Reviews and Recommendations 3
T Design of Experiments - DOE with Centre Point Using Minitab Software 7
Q Design of Experiments Using Subgroup Data - Conducting a 2^3 factorial DoE Quality Tools, Improvement and Analysis 6
T Screening DOE (Design of Experiments) with qualitative response Quality Tools, Improvement and Analysis 23
A The basics of DOE (Design of Experiments) & Validation Statistical Analysis Tools, Techniques and SPC 4
M Design of Experiments (DOE) - Help with homework problem Quality Tools, Improvement and Analysis 5
M DOE (Design of Experiments) with 3 factors with levels of 3,3 and 5 Quality Tools, Improvement and Analysis 3
P What if DOE (Design of Experiments) is not possible? Six Sigma 16
J DOE (Design of Experiments) for ultrasonically welding plastic lids Quality Tools, Improvement and Analysis 15
C Shainin and DOE (Design of Experiments) Information Misc. Quality Assurance and Business Systems Related Topics 2
A DOE - Design of Experiments? What does that mean? Quality Tools, Improvement and Analysis 5
P DOE (Design of Experiments) or Hypothesis test? Applicability and use of each? Six Sigma 6
Tim Folkerts Do you DOE? Use of "Design of Experiments" Quality Tools, Improvement and Analysis 35
F DOE (Design of Experiments) Help: Montgomery's Industrial Engineering book Statistical Analysis Tools, Techniques and SPC 10
A Shainin vs. Fisher vs. Taguchi DOE (Design of Experiments) Statistical Analysis Quality Tools, Improvement and Analysis 10
A DOE - Design of Experiments - Shainin Design and Development of Products and Processes 7
C Seeking Help: Design of Experiments - DOE Statistical Analysis Tools, Techniques and SPC 4
J Selecting factors for screening DoE design Statistical Analysis Tools, Techniques and SPC 8
D Minitab - Design combinations (DOE) Using Minitab Software 4
P Help Setting Up and Analyzing 3 Factor 2 Level Full Factorial Design for DOE Using Minitab Software 3
B Analyze 'Factorial Design' or 'Variability' - Difference when used for DOE Analysis? Using Minitab Software 1
R To Find Out Second Order Model from First Order Factorial Design (DOE) Using Minitab Software 9
B Validation of DoE (Design of Experiements) Model - Minitab 15 Using Minitab Software 3
C DoE and ANOVA - 4 factor and 3 levels Taguchi design with 2 replicates Using Minitab Software 14
Similar threads


















































Top Bottom