The term Industry 4.0 refers to current technological changes in the industrial production environment enabled by information technology advances. The focus of Industry 4.0 is the intelligent factory, i.e., the connection of cyber-physical production systems with Internet of Things (IoT) technology and intelligent data analysis. A core element of Industry 4.0 is the digital twin: a virtual model of a product, the machines, or the production process created with data collected by sensors that enable simulations or real-time analyses of the production status ^[1]. As a digital twin integrates real-time data, it provides a detailed model that can support decision-making through simulation.

The use of digital twins seems beneficial in food processing for various reasons. The Corona pandemic demonstrated the vulnerability of food supply chains and, thus, the need for higher resilience. Production processes must allow increased flexibility and adaptivity to ensure food supply, which requires traceability. The survey “Die Ernährung 4.0-Status Quo, Chancen und Herausforderungen” (Nutrition 4.0-Status Quo, Opportunities, and Challenges) conducted by the digital association Bitkom and the Federation of German Food and Drink Industries showed that 70% of the more than 300 companies surveyed in the food industry consider end-to-end traceability from the origin of the goods to the customer to be a critical scenario for the current decade ^[2]. Various types of sensors exist to support this. However, the potential is far from being exploited. Furthermore, product quality is influenced by different quality levels of input materials. Especially in case of seasonal fluctuations of this raw material quality, an adjustment of parameters in the production process is essential. Introducing new products related to existing ones is also a challenge in food processing. A digital twin of existing products could simplify the introduction processes of new products. The digital twin can learn the correct process parameters for production and is used as a knowledge foundation within a self-adaptive system ^[3]. All those application scenarios show the potential of digital twins in the food supply chain and their huge possibilities, e.g., in determining food quality, traceability, or designing personalized foods.

2. System Model of a Hybrid Digital Food Twin

With the help of machine learning and artificial intelligence, the digital twin is generated from production data and additional data sources (e.g., scientific models, process data, or raw material data) to ensure the traceability of the production and the food status, but also to enable the simulation of the variability of the food in the process operation. Figure 1 shows ourthe concept of the digital twin. In the figure and the following, weresearchers focus on the example of a dairy product (e.g., cheese). The digital twin gets its data from the production site (i.e., sensor, machine, and processing data, e.g., temperature, pressure, or pH value) and also integrates raw material data, complaints, and knowledge from experts (e.g., about the handling of production issues). Using different simulation methods based on chemo-physical models and numerical simulations from food science, the digital twin provides information about food processing and supports real-time feedback on the food process operation. Additionally, those simulations based on scientific models help predict how the product will be changed through the processes and conditions. This information can be used to generate forecasts on how the process steps might influence the quality of the product. Accordingly, the digital twin is suitable for a retrospective but also predictive analytics of the process and product quality. WResearchers want to illustrate the potential of the example of yogurt fermentation. Applying a traditional digital twin concept known from Industry 4.0 would not be feasible. Those concepts use to process data (mainly from machines) to control and describe the production processes. Without actions from the machines, the product’s state will not change. However, for yogurt fermentation, the process is mainly based on resting after inoculation with starter cultures. Hence, the process data cannot sufficiently describe the process as the procedure happens within the product. When weresearchers complement the spare process information with known models from science to describe the behavior of the bacteria, weresearchers get a more accurate picture. But also, the model itself would not be sufficient after inoculation with starter cultures, as the model abstracts and each batch of starter culture also have variations, similar to the milk whose properties differ over the year (due to different feeding). Accordingly, a mixture of both is essential: The model for understanding how the transformation from milk to yogurt works and the available data to adjust the model parameters. This is what weresearchers want to target with ourthe digital food twin concept.