- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Report Inappropriate Content
Which design to choose?
HI,
I want to study the impact of tempertature and speed on the performance of the product.I want to see also their combined effect.
I want to look at three temperatures 120, 140, 160. And three speed 100.150,200. When i do a custom design. its not choosing the mid values. How can I include it. bcos the mid values are the most important ones.
Should i go for a screening design or response surface design. I cannot afford many runs. Can someone suggest a solution.
thanks in advance
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Report Inappropriate Content
Re: Which design to choose?
Hi @Mathej01,
If you only have 2 factors and are interested by middle values, I would recommend using a "traditional" response surface model, like a Central Composite Design. This type of design will also be generated by the Custom Design platform as well when you specify a model with main effects, interactions and quadratic effects (with more flexibility on the runs size).
If you use the Custom Design platform, in order to include middle values in the design, you need to include the quadratic effects of each factor (temperature*temperature and speed*speed). It's also interesting to keep the interaction term in the model if you want to assess any interaction between these two factors (synergistic or antagonist effect) :
Depending on which platform you use, you'll have various possibilities with different runs size, from a minimum of 6 (with Custom Design), to different higher runs size with Classical Response Surface Design platform (DoE, Classical, Response Surface Design) :
You can try to generate several Response Surface Designs with the two platforms and compare the most promising designs with the Compare Designs Platform.
I hope this first answer will help you,
"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Report Inappropriate Content
Re: Which design to choose?
Hi @Mathej01,
If you only have 2 factors and are interested by middle values, I would recommend using a "traditional" response surface model, like a Central Composite Design. This type of design will also be generated by the Custom Design platform as well when you specify a model with main effects, interactions and quadratic effects (with more flexibility on the runs size).
If you use the Custom Design platform, in order to include middle values in the design, you need to include the quadratic effects of each factor (temperature*temperature and speed*speed). It's also interesting to keep the interaction term in the model if you want to assess any interaction between these two factors (synergistic or antagonist effect) :
Depending on which platform you use, you'll have various possibilities with different runs size, from a minimum of 6 (with Custom Design), to different higher runs size with Classical Response Surface Design platform (DoE, Classical, Response Surface Design) :
You can try to generate several Response Surface Designs with the two platforms and compare the most promising designs with the Compare Designs Platform.
I hope this first answer will help you,
"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Report Inappropriate Content
Re: Which design to choose?
Hi,
Thank you @Victor_G .
What about classical main effect screening design?
Will that be an option?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Report Inappropriate Content
Re: Which design to choose?
Hi @Mathej01,
Two-levels screenign designs will only investigate main effects and if possible the interaction between the two factors, with 4 runs minimum. It's a cheap option, but you won't gain a lot if you need after to augment the design, add quadratic effects, and possibly group new runs with a blocking effect. Doing all the experiments in one shot seems affordable with only 2 factors, or you can use a blocking factor with the Custom Design platform if you can run only a very limited number of runs.
With classical two level screening designs, you won't have the middle levels in the design as you intend to have, so no possibility to detect curvature and assess quadratic effects.
Hope this is clearer,
"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)