A model predicts ONE Answer. Industrial reality requires several of them at once: maximizing strength andminimize cycle time and Maintaining a target rating. That is the purpose of this page.
On the left is the list of models for the project: we Check the ones you want to optimize together. On the right, for each one:
- sound type of optimization : Maximize, minimize or target, with its target value in the latter case; ;
- his desirability parameters : the thresholds that convert the response into a satisfaction score ranging from 0 to 1: Worse and Target with a simple goal, Worst Min / Target Min / Target Max / Worst Max for a specific objective; ;
- sound Weight, which highlights its relative importance; ;
- and reading the current point: Prediction, Measured, Error %, Desirability.
The factor sliders move the working point, and all results are recalculated. The calculate button finds the’global optimum : the point that maximizes overall desirability.
Choose from three charts:
- all models,
- the desirability,
- or a specific model.
💡 The desirability is the product of individual desirabilities. This multiplicative form is not merely a computational detail: it means that’A response to 0 cancels everything. A setting that is excellent on four criteria but unacceptable on the fifth is rejected, which is exactly the desired behavior—and something a weighted average does not do.
⚠️ The optimal solution found is not valid only within the data domain. The model readily extends its predictions beyond that point without indicating it. An optimum that lies at the boundary of the domain is a warning sign: the true optimum is likely outside the domain, and reaching it requires further trials rather than extrapolation.