Prediction profiler - parameter operating window

[#1]

Dear all

Hope all are doing well. I have a quick question. I am at the prediction profiler step and JMP has offered the parameter conditions for maximum desirability. Monte Carlo simulation was used to test the sensitivity of the model injecting random variations from the normal distribution. Now, I wanted to know how we can find out the normal operating and maximum acceptable window using the prediction profiler of optimal parameter conditions.

Thanks

[#2]

You continue to interact with the simulation in the Prediction Profiler. I assume that you added the Response Limits column property to your response data column because you were able to maximize the desirability. Consider the variation of each factor. Is it fixed or variable run to run? How much does it vary, if it varies, and in what way (distribution)? What is the expected rate of failure? If it is very low, then you might be able to widen the distribution. If the failure rate is high, then you might tighten the distribution. You have to consider what controls the variation and the cost of control. JMP makes the simulation quick and easy, so you can iterate quickly over different factor settings and ranges to converge on an acceptable outcome.

I realize that this answer is somewhat vague. It highly depends on your situation.

A new tool is coming to the Prediction Profiler in the next version of JMP that was created specifically for this task.

[#3]

Hi Mark Thank you very much for your reply. I have some queries on choosing the distribution. There are several distributions available in prediction profiler, would you please provide an idea what determines to choose a specific distribution over others? Also, would you elaborate on your thought that you mentioned 'Consider the variation of each factor. Is it fixed or variable run to run? How much does it vary, if it varies, and in what way (distribution)?'

Also, what is the best practice to inject the random variations in parameters, is it one at a time or else? Is there any recommendation to follow this? Thanks

[#4]

Choose a distribution that resembles the expected run-to-run variation of the factor level. It models the variation of the inputs to your process. Historical data is excellent for this purpose. Do you record factor levels for each run? Do you plot the levels in a run chart or control chart?

Your second question is a continuation of the first question I already answered.

You are simulating a process run where each factor varies simultaneously, and a random variable is fed through the transfer function to yield the expected output. The regression model is the transfer function. I assume that you have read the Prediction Profiler's documentation for the Simulator feature.