Hi @AnkushSethi,
Welcome in the Community !
Defining the correct assumed model (or a base model adequate to the problem at hand) is not always trivial. Depending on your objective (screening, exploration, optimization, prediction...) , number of factors, experimental budget constraint (number of runs possible), you may have different options. The assumed model helps defining the supposed equation that answer your problem, and helps generating design points that are helpful to determine this model/equation. Note that you don't need to specify perfectly the model at this stage, you will analyze the results and fit a model once you get the results, and the final model can be simpler (with less terms) than the assumed model you have specified during design creation. Also if your assumed model is too simple (missing terms), you can augment your design to explore more complex models with more terms using your preliminary experiments with new generated ones, without starting from scratch. Finally, if you have no idea about the assumed model and/or are not interested in the model form, you can also try space filling designs, designs which will generate points uniformly and spread out in the experimental space, to enable optimization and prediction thanks to various modeling approaches (standard multivariate linear models and Machine Learning models).
You provide very limited context to understand the objective of your problem and the strategy you want to use. I'm also a bit confused by the objective of your response (Maximize or Minimize) and the need to have medium values in the response. You can define an objective as "Match Target" with any value expected for the optimum (medium value or any value).
Nevertheless, if you have low number of factors and are interested in optimizing the response to get medium values, you could specify a Response Surface Model (click on RSM in the Model panel), to ensure factors will be tested at 3 levels: minimum, middle, maximum of the range. This can help to have more variability in the response values, and avoid situations where you have only extreme response values when studying only min/max levels of factors.
If your experimental budget is limited and you can't afford spending runs to test a full RSM model, you can still add centre points in your design (specify number of centre points in Design Generation panel) to test what is the response value in the centre of your experimental design.
I hope I have understood your problem and that these two options may help you,
Victor GUILLER
"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)
