(82a) Using Machine Learning to Develop Deeper Understanding of Agitated Filter Dryer
AIChE Spring Meeting and Global Congress on Process Safety
2019
2019 Spring Meeting and 15th Global Congress on Process Safety
Industry 4.0 Topical Conference
Big Data Analytics and Fundamental Modeling I
Tuesday, April 2, 2019 - 10:15am to 10:45am
Drying is a key unit operation in the manufacture of active pharmaceutical ingredients (API). A
contained environment is necessary for potent APIs, which demands the use of agitated filter
dryers (AFD) for filtration and drying. Understanding of the challenges and mechanism of drying
is extremely important to ensure manufacturing consistency, yet despite being an essential
downstream process in API manufacturing, it is one of the least understood unit operations.
Challenges such as agglomeration and attrition pose threats towards attaining the desired
physical properties of particles. Incomplete understanding and inefficient drying protocols can
significantly affect project timelines and increase the waste of highly costly APIs.
Many experimental and computational methods have been developed to investigate the reason(s)
behind all the bottlenecks, but they are still inadequate due to various factors such as being
unable to deconvolute the complex interplay between the solvent and the powder, process
stochasticisity, scale-dependency at manufacturing level, and computational time and resources.
In this study, we have elegantly combined machine learning with discrete element method
(DEM) and experimental approaches to find innovative solutions and ways to overcome the
aforementioned challenges in order to design a scale-up protocol for the agitated filter dryer.
Some of the key highlights of this integrated approach are prediction of the mixing time required
to reach homogeneity, agglomeration probability, and attrition probability.
Similar methodology can be used to study other complex processes struggling with poor
understanding of mixing, agglomeration, and/or attrition.
contained environment is necessary for potent APIs, which demands the use of agitated filter
dryers (AFD) for filtration and drying. Understanding of the challenges and mechanism of drying
is extremely important to ensure manufacturing consistency, yet despite being an essential
downstream process in API manufacturing, it is one of the least understood unit operations.
Challenges such as agglomeration and attrition pose threats towards attaining the desired
physical properties of particles. Incomplete understanding and inefficient drying protocols can
significantly affect project timelines and increase the waste of highly costly APIs.
Many experimental and computational methods have been developed to investigate the reason(s)
behind all the bottlenecks, but they are still inadequate due to various factors such as being
unable to deconvolute the complex interplay between the solvent and the powder, process
stochasticisity, scale-dependency at manufacturing level, and computational time and resources.
In this study, we have elegantly combined machine learning with discrete element method
(DEM) and experimental approaches to find innovative solutions and ways to overcome the
aforementioned challenges in order to design a scale-up protocol for the agitated filter dryer.
Some of the key highlights of this integrated approach are prediction of the mixing time required
to reach homogeneity, agglomeration probability, and attrition probability.
Similar methodology can be used to study other complex processes struggling with poor
understanding of mixing, agglomeration, and/or attrition.