Bayesian estimation of product attributes from on-line measurements in a dropwise additive manufacturing system

In the pharmaceutical industry, application of additive manufacturing technologies presents unique opportunities for monitoring and control of drug product quality through on-line image acquisition systems which estimate process output from images of emergent drops. For doses produced from pure fluids on-line imaging has demonstrated utility (Hirshfield, et al., 2015); however, processing of particulate suspensions requires consideration of uncertainty regarding gravity- and flow-mediated particle-liquid segregation. Even for well-mixed suspensions, random localization of particles during the necking of the liquid bridge results in variable drop trajectory (Furbank & Morris, 2004), which must be corrected for in the estimates of drop volume provided by an on-line imaging system. This work explores the use of Bayesian statistical models to resolve potential inaccuracy in drop volume estimates obtained by an on-line image sensor in a dropwise manufacturing process for suspension-based drug products. The framework in the case study presents a method by which uncertainty arising from on-line process measurements may be reconciled for through sufficient sampling and off-line analysis of process outputs.