Mapping process-product relations by ensemble regression – a granulation case study

Abstract

Measurement and control of product-process relations are important for particulate products and processes, especially with respect to distributed product specifications, for example size and shape distributions. This paper shares an ensemble regression technique for linking process parameters with distributed granule size and shape characteristics obtained by dynamic image analysis. A case study of fluidized bed granulation is presented, linking process effects (inlet air flow, air temperature, binder spray rate, and binder content) with product attributes. The process was bounded using stable operating criteria defined by balancing wet binder addition with drying enthalpy of the fluidization airflow. Detailed mapping of size and shape trends were obtained using the regression of fully distributed image analysis data obtained from only a few pilot-scale runs, the key enabler being the simultaneous use of distributed size and shape data with process parameters in an ensemble regression model. Results suggest opportunities for optimization linking process and product objectives. More broadly, the methodology suggests opportunities for improved product/process modeling with distributed size and shape data.