Hi All,
I performed RSM in my data with two factors pH and cond. and my response variable is DBC. After conducting the RSM analysis in RSM package of R I got the one stationary points in original units as pH= 5.77 and cond= -2.02. I am unable to understand why I got stationary point of variable cond in minus. I am attaching the output and my data both in original units and coded units.
dear Shantanu,
below i make the following suggestions and observations :
1) In data.frame(8) of your attached data set , at line 7 and 8, the experimental setting is (-1,0) in both, check whether this is a typographical error, if not the design you are using is an unbalanced one, with possible loss of orthogonality among some of the model regressors, if this do not affect your analysis, there is no issue with the design as such.
2) the stationary point in the coded space lie outside the experimental region encompassed by your design, so the response surface that you have, according to the eigen analysis is a negative definite type but a rising ridge kind of surface over the actual experimental space you are dealing with. I suggest that based on your current setting (PH=5, COND=10) you carry the steepest ascent algorithm with a sufficiently small step sizes in both variables, at the experimental setting you find a peak and subsequent decrease in response variable, repeat you design taking the peak point as the design center, i suggest you go for CCD or Box Behnken type of design about this center point, I believe then you will be able to find the optima more accurately than using the current design about the current experimental region you are working with. However, if the objective of your analysis is to develop a Response surface over this pre-chosen Experimental space, your response surface so produced can be used for further analysis of the response behavior over the particular experimental space.
3) A good discussion on different types of response surfaces arising in RSM , is provided in the following book:
Response Surface Methodology: Process and Product Optimization Using Designed Experiments, 4th Edition
Raymond H. Myers, Douglas C. Montgomery, Christine M. Anderson-Cook
4) the numerical computations relating the coded and un-coded regressors, pertaining to your data-set, are okay in my opinion.
hope this helps.
What i referred to as data.frame (8), is actually the data.frame(B) pertaining to your attached dataset.
Hi Debopam Ghosh
Thank you very much for your valuable suggestions.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers