Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Diseño optimo
Este tutorial detalla cómo el so ware Design-Expert ® crea un experimento de método de
superficie de respuesta (RSM) dentro de un espacio de proceso irregular. Si ene prisa por
aprender solo lo esencial del diseño óp mo basado en computadora, omita las secciones de
Notas . Sin embargo, vale la pena el empo dedicado a explorar cosas.
Sin embargo, se ha descubierto que para iniciar la deseable gela nización del almidón, el
empo debe ser de al menos 19 minutos cuando la temperatura está a 110 ° C, el extremo
inferior de su rango experimental. Por otro lado, cuando la temperatura aumenta a 180 ° C, el
almidón se gelificará en solo 17 minutos.
Para resumir: En el nivel más bajo de A, el factor B debe ser al menos 19, mientras que en el
nivel más bajo de B, el factor A debe ser al menos 180. Para complicar aún más las cosas, el
experimentador sospecha que la superficie de respuesta puede ser ondulada. Es decir, el
modelo cuadrá co estándar u lizado para los métodos de superficie de respuesta (RSM)
puede fallar para proporcionar predicciones precisas. Por lo tanto, se recomienda un modelo
cúbico para el diseño.
Un problema como este puede ser manejado por Design-Expert a través de sus herramientas
de restricción y capacidad de diseño óp ma.
file:///C:/Program Files/Stat-Ease/Design-Expert 11/help/tutorials/optimal-numeric.html 1/13
Diseña el experimento
6/12/2019 Diseño óptimo - Documentación Design-Expert 11.1.2
En Diseños estándar, haga clic en Superficie de respuesta para expandir su menú. Seleccione
Óp mo (personalizado) como diseño, e ingrese los límites L [1] (inferior) y L [2] (superior)
como se muestra a con nuación.
En esta etapa, si conoce las restricciones mul lineales, puede ingresarlas directamente
haciendo algunos cálculos matemá cos.
file:///C:/Program Files/Stat-Ease/Design-Expert 11/help/tutorials/optimal-numeric.html 2/13
6/12/2019 Diseño óptimo - Documentación Design-Expert 11.1.2
Nota
Mul factor constraints must be entered as an equa on taking the form of: βL ≤ β1A+
β1B…≤ βU, where βL and βU are lower and upper limits, respec vely. Anderson and
Whitcomb provide guidelines for developing constraint equa ons in Appendix 7A of their
book RSM Simplified. If, as in this case for both A and B, you want factors to exceed their
constraint points (CP), this equa on describes the boundary for the experimental region:
A−LLA B−LLB
1 ≤ +
C PA −LLA C PB −LLB
where LL is the lower level. In this case, lower levels are 110 for factor A and 17 for factor
B. These factors’ CPs are 180 for factor A and 19 for factor B (the endpoints on the
diagonal constraint line shown in the figure below).
Constraint points
A−110 B−17
1 ≤ +
180−110 19−17
However, the so ware requires forma ng as a linear equa on. This is accomplished using
simple arithme c as shown below.
A−110 B−17
1 ≤ +
70 2
1550 ≤ 2A + 70B
Dividing both sides by 2 simplifies the equa on to 775 ≤ A + 35B . This last equa on
can be entered directly, or it can be derived by Design-Expert from the constraint points
while se ng up a RSM design.
Constraint Tool bu on. It’s already set up with the appropriate defaults based on what
you’ve entered so far for the design constraints.
Constraints tool
As directed by the on-screen instruc ons, start by specifying the combina on, or Vertex in
geometric terminology, that must be excluded – shown in red below
Vertex to be excluded
Double-click the cell for vertex A and select 110 from the drop-down menu.
The next step indicated by the instruc ons is to enter in the Constraint Point field the level
for B ( me) which becomes feasible when the temperature is at its low level of 110. Enter 19
– the me need at a minimum to ini ate gela niza on of the starch.
Now we turn our a en on back to factor A (temperature) and its Constraint Point. It must be
at least 180 when B ( me) is at its vertex (low) level.
No ce that Design-Expert changed the inequality to greater than (>), which is correct.
Press OK to see that Design-Expert calculated the same mul linear constraint equa on that
we derived mathema cally.
Note
Take a moment now to read the helpful informa on provided on screen, which provides
some pros and cons to these defaults. Note that we recommend I-op mality for response
surface designs like this. Press screen ps (light-bulb icon) for more details about the
“Op mal Model Selec on Screen.” Feel free to click the link that explains how Design-
Expert uses the “Best” search from coordinate and point exchange.
Press the Edit Model bu on to upgrade the default model of quadra c to Cubic. (Recall that
the experimenter suspects the response might be ‘wavy’ — a surface that may not get fit
adequately with only the default quadra c model.)
The defaults for the runs are a bit more than the experimenter would like to run for lack of fit
and replicates. To limit the number of runs, change the Lack of fit points — from 5 (by default)
to 4. Also decrease the Replicates to 4. The experimenter realizes this will decrease the
robustness of the design, but decides that runs are too costly, so they need to be limited.
Press Next to go to the response entry — leave this at the generic default of “R1” (we will not
look at the experimental results). Then Finish on to build the op mal design for the
constrained process space.
Note
file:///C:/Program Files/Stat-Ease/Design-Expert 11/help/tutorials/optimal-numeric.html 7/13
6/12/2019
At this stage, Design-Expert randomly chooses a set of design points called the
Diseño óptimo - Documentación Design-Expert 11.1.2
Because of the random bootstraps, design builds vary, but all will be essen ally (for all
prac cal purposes) as op mal. Then to see the selec on of points:
You should now see a list of design points that are iden fied and sorted by their geometry.
Due to the random bootstraps, your points may vary from those shown below.
Design layout sorted by ID with point type shown (your design may vary)
Note
Look over the design and, in par cular, the point types labeled as “vertex”: Is the
combina on of (110, 17) for (A,B) excluded? It should be – this is too low a temperature
and too short a me for the starch to gel.
P.S. To see how Design-Expert built your design, click the Summary node under the
Design branch.
To accurately assess whether these combina ons of factors provide sufficient informa on to
fit a cubic model, click the design Evalua on node.
evalua on. Press Results to see how well it designed the experiment.
No aliases are found: That’s a good start! However, you may be surprised at the low power
sta s cs and how they vary by term. This is fairly typical for designs with mul linear
constraints. Therefore, we recommend you use the frac on of design space (FDS) graph as an
alterna ve for assessing the sizing of such experiments. This will come up in just a moment.
Press the Graphs tab. Let’s assume that the experimenter hopes to see differences as small as
two standard devia ons in the response, in other words a signal-to-noise ra o of 2. In the
FDS tool on the le enter a d of 2. Press the Tab key to see the frac on of design space that
achieves this level of precision for predic on (“d” signifying the response difference) given the
standard devia on (“s”) of 1.
FDS graph
Ideally the FDS, reported in the legend at the le of the graph, will be 0.8 or more as it is in
this case. This design is good to go!
Note
A great deal of informa on on FDS is available at your finger ps via the screen ps. Take
a look at this and click the link to the “FDS Graph Tool.” Exit out of the screen- ps help
screen when finished.
On the Graphs Toolbar press Contour to see where your points are located (expect these to
differ somewhat from those shown below).
Our points appear to be well-spaced and the choices for replica on (points labeled “2”) seem
reasonable. Select View, 3D Surface (or choose this off the Graphs Toolbar) to see the
surface for standard error.
This design produces a rela vely flat surface on the standard error plot, that is, it provides
very similar precision for predic ons throughout the experimental region.
This concludes the tutorial on op mal design for response surface methods.
Note
Ahora estás por tu cuenta. Dependiendo de los puntos elegidos por Design-Expert para
su diseño, es posible que desee agregar algunas carreras más * en los vér ces extremos, a
lo largo de un borde, o en otras ubicaciones para disminuir las variaciones de predicción
rela vamente altas. Si modifica el diseño, vuelva atrás y evalúelo. Verifique la vista 3D del
error estándar y vea cuánto más baja se vuelve en el punto (s) que agrega. En cualquier
caso, puede estar seguro de que, a través de su capacidad de diseño óp mo, Design-
Expert proporcionará una buena distribución de las condiciones de procesamiento para
adaptarse al modelo que especifique, y lo hará dentro de la región fac ble que le indique.
* Detallamos cómo modificar el diseño del diseño en la Parte 2 del Tutorial general de un
factor (Caracterís cas avanzadas).