Next Generation Continuous Flow Reactor Technology

To learn more or register for this event, please use the links above or click on the image.

Presentation by Perceptive Engineering

The paper from Perceptive Engineering will describe how the latest thinking on machine learning is being incorporated into the design of advanced process monitoring, control and optimisation tools for pharmaceutical manufacturing processes.

Artificial Intelligence (AI) and Machine Learning (ML) have become ubiquitous terms in the past couple of years. Additionally, the application of this broad range of algorithms to traditional and flow chemistry is often discussed. This is usually limited to problems where a large amount of data is available, something that if often not the case in process development. Futhermore, the question of process control is frequently overlooked during the application of AI and ML to flow processes.

In order to increase industry understanding and uptake of AI, ML and Advanced Process Control, it is necessary to address these gaps and provide clarity in where AI and ML fit within the process control landscape. When we compare “traditional” data-driven techniques with “contemporary” Machine Learning, we see the same algorithms being applied and the common goal of improved decision making through data analysis, prediction and adjustment. There is a difference however, and we’ll explore what it is within the talk.

Content

This presentation inspects and compares the current AI, ML approaches with alternative techniques which are already well established within the process industries, and now also in many of the innovative Pharmaceutical companies.

We will highlight the often-overlooked aspect of data generation and present methodologies to improve this. Furthermore, we will take the process from this initial data generation through a series of steps utilising ML algorithms to produce an optimised and controlled process which is robust and resistant to perturbations.

Case studies from Pharmaceutical processes are used to demonstrate the application of several forms of Machine Learning for process control and optimisation.

Abstract

Advanced Process Control (APC) and Multi-Variate Analysis are data driven techniques to build models, understand the product and process, monitor for abnormal operational events and more recently directly adjust the process to achieve closed loop control of product Critical Quality Attributes (CQAs).

Machine Learning (ML) is often used to describe a myriad of statistical techniques which generate knowledge from data, the key to successful implementation of ML is two-fold. Firstly, the correct algorithms must be selected; secondly, “good” data must be used. Within pharmaceutical process development good data is often not readily available as the process is by its very definition still under development. Quality by Design (QbD) methodologies enable the collection of useful data, however these often require extensive manual experimentation and as such a significant investment in materials and engineering time.

The research presented here shows how careful selection of algorithms from both ML and traditional process control enable the rapid automated generation of useful data which can be used for process control and understanding. This is compared with the traditional QbD and “traditional” data-driven APC techniques.

The result is a pathway for rapid process development, enabling the maximum information to be gleaned from the minimum number of experiments, which yields an optimised and controlled process which is robust and resistant to perturbations.

Case studies from Pharmaceutical processes are used to demonstrate the application of the methodology. By comparing these Machine Learning algorithms with other tools in the process optimisation “toolbox”, we can examine the benefits and challenges to implementing this technology within the industry.