Use the Simulation Experiment option in the Simulator red triangle menu to run a designed experiment to simulate responses across factor combinations. When you select Simulation Experiment, a window appears that enables you to specify various settings of the designed experiment.
The simulation generates a new output table of the space filling design runs with simulated responses for each run. The simulated response columns are the mean and standard deviations of the N simulations per run. Each simulation generates a random value for each design factor drawing from the specified distribution for that factor. The response is then calculated for that simulation run, with the average over all runs reported in the simulation data table.
The Simulation Experiment window enables you to specify the experiment settings.
• The number of experimental runs or design points.
• The number of simulations per experimental run. The default number is determined by the N Runs value specified in the Simulator report.
• The portion of the factor space to be used in the experiment.
• The space-filling design generator. You can choose a Random Culling Near Neighbors design or a Random Latin Hypercube design. The space-filling design is created based on only the active (unlocked) factors.
• If there are specification limits on one or more the responses, there is an option to run a fitting script and optimize the factor settings. This runs a Gaussian Process model and the model report is added to the Profiler report window.
• If factors have a normal distribution specified, then there is an option to substitute a weighted normal distribution for the normal distribution.
Note: If this option is selected, additional runs might be needed for the weighting. This means the number of run reported in the simulation table might be greater than the specified N Runs value.
• The factors to include in the experiment. This can include fixed factors, including categorical fixed factors. Locked factors are shown in the window, but you cannot include them in the experiment. Factors that are not selected for the experiment are set to their current profiler values and shown as a constant column in the output table.
• Shows which responses have random noise added to them.
When you click OK, an output table is created that contains one row for each design point and columns for the design factors and simulated responses.
For responses with specification limits, the simulated responses include the defect rate and an overall defect rate. There is a Gaussian Process script and a Neural script in the output table that you can use to model the overall defect rate from the simulation experiment. In JMP Pro, the Gaussian Process model uses the Fast GASP option. For more information about Gaussian Process models, see “Gaussian Process” in Predictive and Specialized Modeling.
Note: The Simulation Experiment respects linear constraints that are incorporated into the Prediction Profiler.