Sequential Sampling Methods for Finding Classification Boundaries in Engineering Applications

Abstract

Classification boundaries play a crucial role in various chemical engineering applications; however, existing methods of finding them are often inefficient and require a substantial number of experiments. Here, three sequential sampling methods are introduced to improve the accuracy and efficiency of finding classification boundaries with binary data outputs as compared with grid-based sampling methods. Each of the three methods is applied to three distinct scenarios: a simple analytical shower problem, finding a bulk oxidation boundary, and finding a two-phase liquid−gas region. Using our methods, we are able to consistently identify classification boundaries and quantify the boundary uncertainty at a higher accuracy and with fewer experiments compared with grid sampling methods. We achieve this with active-learning-inspired techniques by increasing sampling densities in regions with higher relevance to experimental goals. This is possible through interpretable model informed predictions.