In the contemporary global market the food industry is facing several challenges. Products often need to be manufactured worldwide on different production sites. In case of complex formulations, such as infant formula, it is difficult to ensure the desired functional properties, such as the required physical and chemical stability, when producing on different locations worldwide.
Various factors can play a role in this, including ingredient sourcing from different providers, differences between production lines and local circumstances such as the climate. Upscaling, manufacturing and product quality problems that arise are mostly solved case-by-case using money- and time-consuming trial-and-error methods rather than a structured approach. Predictive modelling, through model-based expert systems, can help you resolve these issues by making optimal use of available knowledge.
Even though knowledge about the effect of ingredients, process design, process settings and storage conditions on product functionality is often available, it is usually not available in the right place and format to ensure effective use.
First of all, knowledge is often only available in a qualitative format, which is definitely useful but insufficient when it comes to preventing problems and trial-and-error work. Quantitative information in the form of experimental data from lab or pilot scale trials are often difficult to extrapolate to larger scale without modelling tools. Even though the availability of large-scale production data has grown exponentially due to the increased use of on-line sensors and data acquisition systems, appropriate tools to interpret these data are often not available. The latter case is also related to the fact that many factors are at play that affect product functionality. A complete set of required knowledge cannot be found with one person or even within one area of expertise: the required knowledge is scattered and cannot be coupled easily.
To meet the production challenges mentioned above you need a system that incorporates integrated, quantitative, extrapolatable knowledge. Predictive models can be a solution, provided that these models are implemented in a suitable software architecture: a model-based expert system.
A model-based expert system is more than a collection of predictive models. First of all, the architecture of the system must allow for coupling of models, preferably in the form of a dynamic flowsheet. The system also has to be accessible to non-experts and suitable for different purposes such as product development, up-scaling and process control. Besides software architectural requirements this also means appropriate user interfaces have to be available. Last but not least, the software architecture should be such that the expert system can easily be extended with new or improved models or additional data (by coupling the system with external databases, data acquisition systems or LIMS).
Developing an expert system involves several steps. The first step is knowledge mapping: based on existing knowledge, all factors that play a role in determining relevant functional properties of the end product are mapped and the interrelationship between these factors is described, preferably in terms of mechanistic understanding of the system. The second step is data collection: based on the knowledge map a list of required data can be compiled. These data are then collected from literature, experimental or production data. The next is developing the sub-models which are integrated into an expert system, as described above, in the final step.
To develop a platform with a suitable architecture Nizo has joined forces with Process Systems Enterprise, developer and provider of the gPROMS software. In particular, the equation-oriented approach, in which the flowsheet is treated as a set of equations to be solved simultaneously, makes gPROMS a suitable basis for model-based expert systems. By integrating our Premia models in the gPROMS environment with a suitable underlying architecture for food production processes, a platform will be developed that is suitable for developing model based expert systems that will meet all requirements.
Our existing software platform called Premia includes predictive models for various processes such as heat treatment, membrane separation, evaporation, drying and cheese manufacturing. Even though Premia uses a sequential modular approach instead of an equation oriented approach, the models have been implemented in such a way that they can be coupled into an expert system. A good example is the gouda cheese model-based expert system. This expert system comprises a set of physico-chemical and statistical models for various parts of the cheese manufacturing and ripening process. Besides product parameters such as moisture and salt content and cheese yield the expert system also predicts parameters such as protein breakdown (proteolysis) and risks for defects such as bitter as a function of processing and ripening conditions.
A similar approach can for example be used to predict sorption isotherms and stickiness of infant formula or other food powders as a function of product composition. Such a system can be used to, for example, predict the fouling behaviour and capacity limitations of spray dryers and changes in powder functionality during storage, taking into account local circumstances such as the local climate. This also enable producers to look at potential bottlenecks in up-scaling and relevant economic parameters such as capacity limitations for specific production locations already at the early stages of product development.
© FoodBev Media Ltd 2020