How can I perform a inverse prediction across several models according to their specification limits

[#1]

Dear JMP community,

I need your help. Based on a DoE analysis I have investigated the impact of 6 input-parameters/factors, their interaction effects and quadratic effects on 30 different output-parameters. Hence, I have created 30 models in the fit model plattform. 

Each of these 30 output-parameters has certain specification limits. I have to evaluate the range for each input-parameter which leads to output values falling within the specification limits while keeping all the other input-parameters on certain conditions. The inverse prediction function in JMP is a big help. However, performing all these steps manually (for each input-parameter and across 30 different models) takes too much time (30*6 = 180). Is there a way to automate this procedure (maybe in JSL?)

All the best!

[#2 · Accepted solution]

I think you could use the profiler for this.  You might try:

• In Column Info, set the response limits for all output columns
• Save the prediction formula for all models from fit model if you did the before setting response limits, you will need to set them here instead.
• If you have a lot of limits including low and high limits on variables, you might make a single column that is true if every column is within limits.
• Open the profiler from the graph menu, add all predicted columns and the limit column if you made one to the y variables
• Enable desirability functions and make sure they match your specs
• Optimize desirability

I know this isn't the same platform as you are using, but this talk gives an example of using the profiler in a similar way.  You can do the same thing with just saved prediction formulas from the fit model platform, you just might need to add 

[#3 · Accepted solution]

I really like your idea. Especially I think I can use it for another probleme I currently facing. However JMP only allows to calculate CI but not prediction intervals. To circumvent this issue, I adapt the prediction SE formula and plot again everything in the profiler with the adapted SE formula. The problem I‘m facing there is that the option inverse prediction is no longer available.
Using your approach allows me only to consider the mean prediction of the model but not the uncertainty of the model.
Maybe you or someone else has another good idea to circumvent that issue?