Talk Back: Quality through quantification: how understanding generates profit.

As a relative newcomer to the pharma sector, and a chemical engineer, one of the things that things always strikes me is how little use there is of the traditional process engineering quantification tools. This is despite the fact that pharmaceutical manufacturers deal with some of the most complex phenomena in the process industries, and manufacture is highly regulated.

Of course PAT is rapidly changing all of this, by driving a need for greater process understanding through analysis and quantification, and ultimately, the development of predictive capabilities. The key to quality and consistency is to understand how things work, and to capture that knowledge in such a way that you can use it to predict behaviour and thus make informed judgments about how to design and operate.

Advanced Process Modeling

A key technology here is Advanced Process Modeling, or APM. There are two magic ingredients for an APM which make it different from the ‘black-box’ simulation models of the past. The first is a detailed first-principles representation of all the relationships that define the process, from the micro-scale diffusion of molecules to the macro-scale circulation of cooling water in a jacket. This is constructed within a modeling framework, by describing the physics and chemistry in terms of heat and material flows, reaction, diffusion, geometry of equipment, interaction between process flow sheet units, and so on. Most of the relationships are in fact well known and well documented, and there is a growing body of well-proven models available, serving to minimize the investment required.

Even when working with fundamental relationships, there are always empirical constants (or model parameters) that need to be determined. This is the second ingredient: accurate parameter information derived from real-world – laboratory, pilot or operating – data, using mathematical optimization-based parameter estimation techniques in a process known as model validation.

It is the combination of these two that provides such models with a highly accurate predictive capability over a wide range of operating conditions. Once you have a fully validated model, the sky is the limit. You can optimize the size and numerous other attributes of reactors; you can optimize recipes and operating policy to reduce batch times and maximize quality; you can scale-up crystallizers secure in the knowledge that they will actually produce crystals of the right size and shape; you can determine the cooling profiles that give the optimal crystal size distribution; you can use the model within the automation framework to monitor batches using inferential measurement, or to generate linearised models for batch model-predictive control; and many others.

Modeling and experimentation

It is easy see how experimental data can be used to improve models, but less well known is that models can be used to improve experimentation with great effect. Build a relatively simple model of the experimental rig, complete with relationships describing the reactions you are studying, for example, and you have a tool that allows you to do many things never previously possible. Model-based data analysis will tell you immediately whether you need additional data in certain areas – for example, if the available data are not sufficient to determine the values for two parameters independently of each other. Using the same model in a model-based experiment design framework will enable you to design the optimal subsequent experiment – that is, the experiment that will yield the most information within the least time, or at the lowest ingredient cost.

This is the basis of a set of techniques called model-based innovation, which are helping to drive change through the chemical and other process sectors. It is also fundamental to another exciting new area, model-based risk management.

Model-based risk management

Model-based data analysis may tell you that a parameter value is likely to be, say, 2.5, but the confidence analysis of the data fit says that the real value could lie anywhere between 2.4 and 3.2. What if the true value is actually 2.4 or 3.2? Will your downstream equipment cope? Will you achieve required conversion and quality? If the answer is “no”, you know that you need to invest in refining the data. This may involve doing a few more experiments, or it may involve building an entire R&D facility. The key here is that the model provides you with a quantification tool that allows you to support risk decisions with hard numbers.

APM has been applied in recent years to great effect in the chemical industries. Pharmaceutical manufacture, with its equally complex reaction, crystallization and biotechnology processes, stands to gain just as much, if not more – particularly when the time-to-market imperatives and the necessity to change rapidly are taken into account.

“It is easy see how experimental data can be used to improve models, but less well known is that models can be used to improve experimentatio