Process Analytical Technology for the food industry

In recent years a wide-reaching recognition of the importance of production consistency and quality has emerged in the food industry. With the recent recognition by the American Food and Drug Administration (FDA), Process Analytical Technology (PAT) has become the key issue in statistical process design, monitoring and control for pharmaceutical industries worldwide1.

By endorsing PAT, FDA aims to create a set of scientific principles and chemometric tools with a strategy for regulatory implementation to accommodate process monitoring and control innovations2. In addition to researching the questions shared by different industrial categories, a crucial part of the successful implementation of in-process monitoring is that each field of application requires solutions to its own specific problems. Several challenges distinguish the food industry from fields with a longer tradition in this area (such as the petrochemical world):

• A largely batch-wise or semi-continuous mode of most unit operations
• Large (biological) variations in raw materials and process feed-stocks
• From a physical, chemical or biological perspective, food products and processes are complex and heterogeneous multi-factorial systems
• The ultimate end-quality evaluation of food products – i.e. that performed by the consumer – can create a taxing optimisation parameter for the control strategy

This article will present our interpretation of PAT’s potential in the food industry by highlighting some of the essential building blocks and opportunities.

Process Analytical Chemistry

Closely linked with the concepts covered by PAT is the more established and continually evolving field of Process Analytical Chemistry (PAC). The role of PAC can be broadly defined as retrieving qualitative and quantitative information from a process3. PAC distinguishes itself from off-line laboratory measurements in that the measurement system is adapted to the process, instead of a sample being tailored for the laboratory equipment. Most food manufacturing processes consist of a series of unit operations each intended to modulate certain properties of the materials being processed. As a consequence of classical sample preparation for off-line/laboratory quality analysis, valuable information relevant for the formulation matrix is frequently lost. Several new technologies are now available that can acquire information on multiple attributes with minimal or no sample preparation. They provide an opportunity to assess several elements instantaneously. Non-destructive sensor-based measurements can thus provide a useful process signature that may be related to the underlying process steps or transformations. Spectroscopy, in the form of on-line analysers for chemical, physical and rheological elucidation, is the main candidate for gathering information in the complex process streams in food material processing and production. However, many competing in-process measurements have been developed, or are being developed and making the right choice is a crucial part of the PAC discipline4.

SPD, SPM and SPC

In its guidance, the FDA states the following definition: “The Agency considers PAT to be a system for designing, analysing and controlling manufacturing through timely measurements (i.e., during processing) of critical quality and performance attributes of raw and in-process materials and processes, with the goal of ensuring final product quality”1. In a free interpretation of this definition we identify three pillars: Statistical Processes Design (SPD); Statistical Process Monitoring (SPM) and Statistical Process Control (SPC).

The goal of SPD, and Design Of Experiments (DOE) in general, is to gather as much knowledge as possible about a process or system from a limited number of experiments. The ultimate aim is often termed ‘Quality By Design’. This involves identifying all uncertainties from inputs or (external) disturbances during development, thus creating robust system designs with little or no surprises during up-scaling and operation. A further important area in SPD is ‘data mining’, where patterns in historic process data are explored, to identify the critical points and disturbances in a production. It is important to mention that we view PAT as a broad framework. An example might include rapid screening of natural/harvesting products by spectroscopic methods; the exploratory classification of the data by chemometrics and a visual (rather than computational/inferential) interpretation.

SPM can be broadly defined as keeping track of the performance and operation of a production system, while SPC would involve a corrective action based on measurements when the process is deviating from a predefined target. Much research on Multivariate Statistical Process Monitoring (MSPC) for batch processes, combining a multitude of process signals such as temperature, flows, pressures, sensors, etc. at model inputs, has been conducted in the field of chemometrics5. In the control area novel run-time optimisation strategies, based on process sensor information and chemometrics, are being developed that regulate process parameters to counter feedstock variations in real-time. An essential feature of the new PAT paradigms is that many statistical devices are based on so-called process signatures or latent phenomena from the field of chemometrics, rather than hard engineering units. Food systems are, by nature, multivariate complex systems and observing one process parameter at a time is insufficient to get the whole picture.

Care must be taken not to interpret the concept of four subjects – PAC, SPD, SPM and SPC – too rigidly. For example, the temporary placement of more sophisticated in-process sensors (sometimes labeled SWAT analysis) to collect data for system identification or troubleshooting could be categorised under all four headers. A successful PAT project will likely include a little from each area. The above description should also make clear that PAT implementations are, by definition, multi-disciplinarian: chemistry for food technologists; process development, equipment design and control infrastructure for engineers; mathematics, statistics and chemometrics for data analysts; sensor selection and reference analysis for analytical chemists, etc. In this respect, any successful implementation of PAT requires the elimination of the ‘islands-of-expertise’ mentality in both industry and academia6.

Future

An enormous number of on-line/in-process sensors are implemented and used today in the global food industry. For example, near infrared spectroscopy (NIR) owes a large part of its present popularity to R&D conducted in the food branch. All these applications – past and present – would fit well under our definition of the PAT umbrella. However, great opportunities for the future are anticipated when the FDA’s PAT concept is embraced by the food industry. The following are examples of possible research areas:

Closely linked with the performance of a monitoring and control system, is the measurement quality. Issues such as probe or window fouling and temperature influences on the measuring system are typical problems and inconsistencies introduced by process conditions in food manufacturing. They distinguish in-process spectroscopy from the well conditioned laboratory world. The development of sophisticated (multivariate, multi-way) sensors and new, robust chemometric calibration methods to deal with such problems, is still very active.

The combination of dynamic models and in-process measurements is another wide research field in food science. A kinetic model – either based on a first principles system or modeled via identified systems built using SWAT analysis as training data – can be used to predict into the future7. The final output of a batch process, for example, is a function of the intrinsic properties of the system (the dynamic model), the feed or charge and the system settings and disturbances during processing. Estimates of the unknowns (inputs and disturbances) can be found during processing from the in-process measurements. This combination of process model and measurements can be used to make endpoint predictions during processing, which gives another view of SPM. Both system model and measurements are imperfect inputs requiring statistics and error propagation to be included in prediction. The research question, then, is where and when to make which measurement during the process, to minimise uncertainty?

An improved process design and control will also lead to new business opportunities by, for example, connecting process operation regimes to specific consumer preferences. An example could be the manufacturing of region-specific products in one single process unit (high-homogeneity products for the Asian market versus well controlled variations over limited range for the European market) by run-time optimisation of buyer demands

We expect a substantially increasing awareness and demand for Statistical Process Design and Monitoring and Control in the future of the food industry. We also anticipate a high integration of the PAT-principles with quality assurance in the industry. In recognition of these expectations a new master program is in preparation for 2006 (working title FoodPAT) at the KVL, Denmark. In the program we incorporate all relevant aspects (food technology, process interfacing, instrumental methods, data analysis, etc.) to prepare the students for the multi-functional challenges ahead.

References

1. “Guidance for Industry, PAT – A Framework for Innovative Pharmaceutical Manufacturing and Quality Assurance, Draft Guidance” available form www.fda.gov/cder/OPS/PAT.htm
2. A.M.C. Davies “What is PAT?” Spectroscopy Europe April/May(2004)33-34
3. J.B. Callis, D.L. Illman and B.R. Kowalski “Process Analytical Chemistry” Analytical Chemistry 59(1987)624A-637A
4. A.K. Smilde, F.W.J. van den Berg and H.C.J. Hoefsloot “How to choose the right process analyzer” Analytical Chemistry 74/13(2002)368A-373A
5. Th. Kourti and J. MacGregor “Tutorial: Process Analysis, monitoring and diagnosis, using multivariate projection methods” Chemometrics and Intelligent Laboratory Systems 28(1995)3-21
6. J.R. Davis and J. Wasynczuk “The Four Steps of PAT Implementation” Pharmaceutical Engineering January/February(2005)10-22
7. F.W.J. van den Berg “Optimal Process Analyzer Selection and Positioning for Plant-Wide Monitoring” Ph.D. Thesis University of Amsterdam (2001) available from www.models.kvl.dk