Speciality Chemicals Magazine SEPT / OCT 2022 | Page 17

PHARMACEUTICALS ratio at plant-scale . A new type of cryostat had to be developed to realise this task .) The reaction kinetics of the chemical transformations were evaluated in the dedicated lab reactors and fitted to the computational model . Additional algorithms described chemical processes including all of the side reactions . This related to prediction models used for the scale-up . We can now simulate changes in process parameters , filling degree or , for example , catalyst performance without additional experiments or tests at scale . The influence of these variations on purity profile , temperature control , turnover rates and heat control are well understood , so that the scale-up from lab ( 1L ) to plant ( 4,000L ) was successful . The scale-up risks for quality and process safety are significantly reduced . Due to the prediction of the
References :
1 : C . J . Welch et al ., Org . Process Res . Dev . 2017 , 21 , 414- 419 . doi : 10.1021 / acs . oprd . 6b00427
2 : T . Furrer , B . Müller , C . Hasler , B . Berger , M . Levis & A . Zogg , Chimia 2021 75 ( 11 ), 948-956 . doi : 10.2533 / chimia . 2021.948 effective reaction time , equipment usage is optimised . Fewer IPCs are needed . Fail batches can be avoided from scratch . The final process can be performed in a robust set of process parameters based on computational predictions . In general we have seen that predicted and real on-scale measured data are fitting very well with the new algorithms .
Physical operations toolbox
Crystallisation , filtration , centrifugation and drying are also of very high interest in the pharmaceutical and fine chemical industries . 1 Because downstream drying is very often influenced by upstream crystallisation and filtration performance , we decided to combine all involved unit operations in one ' physical operation toolbox '. Classically , crystallisations are controlled by controlling temperature profiles , dosing rates , seeding processes , etc . Well-established strategies to control oversaturation , crystal growth and nucleation make it possible to influence the final particle size distribution ( PSD ) of the crystals reliably . Prediction models are commercially available and are part of our toolbox . However , some crystallisations and precipitations are very challenging . Typical reasons are : ( semi ) - amorphous materials , very small PSDs and the quality or composition of the raw product interfering with crystal growth or nucleation . To isolate the materials , it may be necessary to find conditions that accommodate an effective agglomeration , so that standard filtration or centrifugation equipment can be used . The properties of the filter cake , especially the amount and composition of residual solvents , have a significant influence on the drying process . Products may start to ball and form lumps , which then cannot be dried effectively . Sticky phases may occur during drying and the product may adhere to the stirrer .
Standard computational models do not reflect these cases . We had to establish a strategy to characterise the specific material properties from crystallisation , centrifugation and drying . This allowed us to study and predict their influence on the downstream process over multiple unit operations . Our toolbox is based on lab programmes for each individual process step , including adjusted computational models , analytical characterisation and , for example , rheological investigations for solidstate properties . The interface from one unit operation to the next is included in the toolbox , so that the influence of a crystallisation parameter on the drying performance and final solid property is well understood . Equipment properties also play a significant role on these processes . So , again , digital and down-scale twins of the production equipment are needed .
Conclusion
The two examples above illustrate the toolbox approach of Siegfried . We have established additional toolboxes for process safety , simple mixing effects , physical unit operations and reaction kinetics . Commercially available computational models are often the basis but they need to be adjusted and accompanied by smart lab programmes to ensure data integrity and significance for the prediction . Based on this approach , we are able to predict process performance on a large scale , even without a piloting step . The prediction enables us to choose the most robust and best process parameters for QbD . •
Dr Michael Levis
HEAD - PROCESS TECHNOLOGIES ( R & D )
SIEGFRIED AG k + 41 62 746 1259 J michael . levis @ siegfried . ch j www . siegfried . ch
SEP / OCT 2022 SPECCHEMONLINE . COM
17