Purpose
Bayesian optimisation proposes the next process setting to be tested, so that the optimum is reached with a minimum number of experiments.
Method
A Gaussian process models the relationship between process parameter x and result y, including its uncertainty. An acquisition function weighs expected performance against uncertainty and proposes the next setting. The model is updated after each measurement.
Optimisation workflow
| Next Process Parameter Setting | – |
|---|
AI Model for Process Optimization
Orange: completed experiments. Blue line: predicted result. Shaded band: 95% uncertainty. Black line: next setting.
Interpretation
A narrow band indicates high confidence near existing measurements; a wide band indicates limited information. Each measurement reduces the uncertainty around its setting. With "Reveal hidden function", the dashed line shows the true process.
Determine Next Setting
Score of each possible setting. The setting with the highest score is tested next.
Interpretation
High scores occur where a good result is expected or where uncertainty is high. Only the position of the maximum is relevant, not the absolute values.
Best result so far
Line: best result after each experiment. Grey: initial experiments. Orange: AI suggestions.
Experiment history
All experiments, newest first. The best result is highlighted.