CLD-to-PSD model to predict bimodal distributions and changes in modality and particle morphology

[Display omitted] •ImprovedCLD-to-PSD model predicts bimodal distributions.•New model wrapper infers crystallization morphology and modality changes.•Different CLD modalities and higher-order moments are used to improve predictions.•Regularization techniques increase model robustness under limited data conditions. This work presents a model architecture which allows computation of the particle size distribution (PSD) from chord length distribution (CLD) measurements. The model is a refinement based on a previously introduced data-driven model to determine 1D/2D PSDs. The initial model was limited to the prediction of unimodal PSDs but enabled robust quantification of CLDs using a small dataset. In this work the previous model is expanded in two directions: first, the model architecture is refined to further increase model robustness and enable the prediction of additional PSD modalities such as bimodal distributions. Second, a meta-architecture is proposed in which an inference system can characterize transitions in the morphology and modality of a particle population using the CLD input signal. This inference system can be trained to detect morphology or modality transitions in crystallization processes.