Issue link: https://www.e-digitaleditions.com/i/1499609
PharmTech.com Trends in Manufacturing eBook May 2023 Pharmaceutical Technology ® 31 Process control dent that our thorough validation process will ad- dress these concerns," she explains. Although the lack of specific regulations for AI is a challenge for the industr y, Gastone points to cur- rent efforts to develop frameworks for the use of AI in regulated environments. FDA, for example, pub- lished a discussion paper in March 2023 requesting comments about AI in drug manufacturing (5). "Additionally, industry organizations such as the Pistoia Alliance are developing best practices and g uidelines for the use of AI in pharmaceutical re- search and development," says Gastone. "We believe t hat t he adoption of AI in t he phar- maceutical industr y can be achieved without com- promising reg ulator y compliance," says Gastone. "By work i ng c losely w it h reg u l ator y bod ies a nd fol low i ng best prac t ices, we ca n ensure t hat our AI-based systems are effective, safe, and compliant with industr y standards." References 1. Luperini Group. Cognitive Pharma Machinery: The Revolution in the Pharmaceutical Industry Begins. Press Release. Feb. 2, 2023. 2. Eigengra n. T he Cha l lenge of Wet Gra nulat ion. eigengran.it. 3. US CFR Title 21, Part 11 (Government Printing Of- fice, Washington, DC). 4. ISPE, GAMP 5: A Risk-Based Approach to Compliant GxP Computerized Systems, 2nd edition; July 2022. 5. FDA. Discussion Paper: A r t i f icia l Intel l igence i n D r u g M a nu f a c t u r i n g , Not ice; Re q ue s t f or I n f or m at ion a nd Com ment s. Fede ral R egi ste r. March 1, 2023. ■ Refining Process Control in Fluid-Bed Granulation Process control capabilities for fluid-bed granula- tion continue to advance. Advanced dynamic process control can account for variations in raw material or atmospheric conditions using data from process an- alytical technology (PAT) to enable automated feed- back control. Innopharma's advanced dynamic pro- cess control (ADPC) system receives real-time data from process sensors, near infrared spectroscopy, and particle size and shape imaging PAT to respond to process deviations and to detect phase changes and endpoints, improving in-process control and reducing batch-to-batch variation (1). Innopharma's SmartX ADPC is "intelligent" because it allows automated feedback process control as well as the potential for predictive, feed-forward control. Rather than being driven by artificial intelligence (AI) or neural networks, however, it uses modular, process-specific logic content blocks based on expert understanding of the processes, explains Chris O'Callaghan, head of engineering at Innopharma Technology. The control system is designed for good manufacturing practice (GMP) processes and is currently being used in development projects. Validation for GMP manufacturing is a hurdle to adoption because this type of control technology is relatively new, but it has been demonstrated to reduce risk (e.g., of over-drying or under-drying), says O'Callaghan. The technology continues to be developed as PAT capabilities grow, and there are opportunities for AI, as well. "AI [enables] fine-tuning, to control the 20–30% of the process that can't be controlled with traditional approaches based on current process understanding and PAT capabilities or to correct for some of the error that is impractical to define," explains O'Callaghan. Innopharma has also used AI, along with traditional approaches, to build simulations of manufacturing equipment, such as fluid-bed granulation with the ADPC controller. "Machine learning, which is a type of AI, is used to solve some of the tricky problems in simulation," explains Claude Lacey, principal software architect at Innopharma Technology. "Machine learning can be used, for example, to model how a heat exchanger will behave in the system, because it can capture the lag from the heat exchanger into the next unit operation. We capture real-world data with experiments and use this to train the simulation." Innopharma is also using AI to boost the abilities of the image analysis software used with its Eyecon 2 on-line particle size and shape analyzer; EyePass software with AI capabilities was introduced in 2022. "Our original particle sizing technology could detect white material on a dark background using traditional imaging techniques and algorithms," says Lacey. "To analyze different colors or different shapes, however, we needed new algorithms. Image processing using deep learning with convolutional neural networks is quite advanced, and we used this type of AI to train models that work on a broad range of sizes and colors." If the generic models are not accurate enough for a specific material, a new model can be trained using additional data. The software can now be quickly adapted to handle new materials. Reference 1. O'Callaghan C. et al., Self-Guided Control of a Fluid Bed Granulation Process. Pharm. Tech. 2020, 44 (1) 42-45. —Jennifer Markarian