Fig. 7. Proposed workflow for integrating hybrid modelling into crystallisation process development and the literature, spanning experimental design, data acquisition, model construction, and multi-scale validation, with optional model comparison between data driven, mechanistic and hybrid models. * A shared design space allows mechanistic and data-driven models to be fitted simultaneously, with parameters adaptable to continuous or mixed strategies as development progresses. ** For example, time-series particle size data may be available without direct kinetic measurements. *** Models can be used to fill data gaps identified in box 2; for example, a PBM-derived mass balance can generate synthetic training data where experimental high performance liquid chromatography (HPLC) data are limited.
