Wed
27
May
Seminars and Conferences
Data-Driven Multivariate Optimization of CFD–PBM Precipitation Kinetic Models for Prototype Design and Scale-Up
On Wednesday, 27 May 2026, will take place an application seminar titled Data-Driven Multivariate Optimization of CFD–PBM Precipitation Kinetic Models for Prototype Design and Scale-Up, within the EMJM Multiphase programme.
Abstract
High-fidelity CFD–PBM simulations can predict reactive precipitation, but their practical value hinges on robust identification of kinetic parameters under multivariate operating conditions. Building on recent work on Mg(OH)₂ precipitation, where nucleation, growth and aggregation kinetics were inferred from particle size distribution data and shown to be transferable across mixer/reactor configurations, this seminar presents a scalable optimization workflow that couples 3D CFD with population balance models for parameter estimation, model discrimination and uncertainty-aware calibration. The approach combines physics-inspired reduced decision variables, surrogate-assisted global search and data-driven strategies (including deep-learning-based multivariate optimization) to efficiently explore correlated parameters and experimental conditions. Once calibrated, the kinetic model is deployed in two directions: (i) guiding the design and operating window of an industrially relevant precipitation prototype via 3D CFD–PBM “digital twin” simulations, and (ii) validating and generating compartment-based reduced-order models to enable fast design iterations and future real-time applications. The talk highlights lessons learned on identifiability, transferability and the trade-off between fidelity and computational cost.
Speaker: Antonello Raponi (EMSE)
Abstract
High-fidelity CFD–PBM simulations can predict reactive precipitation, but their practical value hinges on robust identification of kinetic parameters under multivariate operating conditions. Building on recent work on Mg(OH)₂ precipitation, where nucleation, growth and aggregation kinetics were inferred from particle size distribution data and shown to be transferable across mixer/reactor configurations, this seminar presents a scalable optimization workflow that couples 3D CFD with population balance models for parameter estimation, model discrimination and uncertainty-aware calibration. The approach combines physics-inspired reduced decision variables, surrogate-assisted global search and data-driven strategies (including deep-learning-based multivariate optimization) to efficiently explore correlated parameters and experimental conditions. Once calibrated, the kinetic model is deployed in two directions: (i) guiding the design and operating window of an industrially relevant precipitation prototype via 3D CFD–PBM “digital twin” simulations, and (ii) validating and generating compartment-based reduced-order models to enable fast design iterations and future real-time applications. The talk highlights lessons learned on identifiability, transferability and the trade-off between fidelity and computational cost.
Speaker: Antonello Raponi (EMSE)