topic Re: Simulations in Discussions

Simulations

CNU — Fri, 09 Jun 2023 00:49:09 GMT

Hi
How do I create a simulation with a relative standard deviation for the input factors?

Re: Simulations

Mark_Bailey — Mon, 09 May 2022 11:43:42 GMT

You can't, if I understand your question. You are definition sampling distributions for the input variables or factors. The standard deviation is in terms of the variable, using the same units. You can convert a relative measure, such as % CV, by multiplying the RSD by the mean.

I might have completely misunderstood your question.

Re: Simulations

CNU — Wed, 11 May 2022 09:01:34 GMT

What I want to do is to estimate the defect rate where one of the input variables is the mass with a relative standard deviation of +/- 2%. The input range for the mass is 20-100 kg.

Re: Simulations

Mark_Bailey — Wed, 11 May 2022 12:44:51 GMT

This simulator assumes that each factor has a constant set point but the actual setting varies from run to run. You provided a factor range of 20 to 100 kg. What is the desired or optimal set point weight?

The RSD is a constant, so the SD is not a constant. You can use a JSL expression to define the distribution of input factor values instead of selecting one of the distribution models with a constant mean and SD.

Re: Simulations

CNU — Thu, 12 May 2022 12:30:20 GMT

Can you help with this JSL expression? How do I define SD from the actual setting and RSD?

Re: Simulations

Mark_Bailey — Thu, 12 May 2022 14:28:38 GMT

We are going in circles. The Profiler uses a fixed factor set point (e.g., W/T = 0.55 above ). It uses this value as the location of the chosen distribution, and it does not change during the simulation. It uses the spread (e.g., W/T SD = 0.03 above) to generate random deviates based on a selected distribution (e.g., normal distribution). If you have a RSD, say 5%, then the absolute SD is RSD * Location = 0.05 * 0.55 = 0.0275.

The choice of Expression allows you to use any valid JSL expression that returns a number. For example:

Random Normal( 50, 5 );

Will generate normally distributed deviates from a normal population with a location of 50 and a spread of 5. You could replace either the constant location or the constant spread or both with another expression that varies the value deterministically or randomly. I do not know how you would use this ability because you still have not answered my previous questions.

See the documentation for the Profiler Simulator factor settings.

Re: Simulations

CNU — Mon, 16 May 2022 11:23:50 GMT

Sorry for not being clear on this.

If I use a fixed set point of W/T = 0.55 and then apply the expression as shown:

Then SD is 0.55*0.05 = 0.0275.
Now, I want to simulate the defect rate applying the sequence location in the range from 0.2 to 0.6. When using the expression will JMP use a constant SD of 0.0275 for each the different locations or will the expression be interpreted as location*0.05?

Re: Simulations

Mark_Bailey — Mon, 16 May 2022 13:55:34 GMT

I do not see a way to make the SD depend on the location with the JMP controls, even with an expression. There is no variable to access the current location or a way to specify a variable SD. You can sequence location or spread, but not both.

You can manually create a table of 25 combinations of W/T, C, and the associated SD, and then manually set the factor location and SD for each run. This approach could also be scripted as the Profiler and Simulator are fully scriptable.