Variations in raw materials and processing parameters are a common challenge in food production, requiring more time and resources to be spent on quality control, food safety and traceability. DT-OptiDry digital twin is designed to overcome these issues by capturing input variability in a virtual environment and simulating the process.

In practice, DT-OptiDry will deliver a calibrated system for controlling raw material and processing variability and adapting the process to achieve the desired output. Manufacturers can then reduce uncertainty related to final products, improve energy efficiency and sustainability and optimise processing time. Overall, they will benefit from a robust system for food safety and quality control.

Lead SME UpIntelligence (Virtual Intelligence S.L.) has partnered with Embutidos Maybe, Spanish manufacturer of traditional sausages, for this validation project.

Final report summary
Data mining techniques were used to analyse product variables (water activity and pH) and environmental conditions at Embutidos Maybe and identify patterns and correlations suitable for process control optimisation and quality management. Water activity was confirmed as the target variable for monitoring the sausage smoking process. Datasets of 20,000-30,000 environmental measurements and more than 400 product samples were used for the development of predictive models.

Machine learning algorithms were developed using production process dynamics to create a smart tool for predictive and simulation purposes. The development of predictive models was based on two main requisites:

1. The solution should be able to predict when the smoking of a product batch is complete. This requires advance knowledge of water activity.
2. To develop the digital twin, the user must be able to modify the data that feeds the predictive system, based on the modelling of water activity.

The digital twin is a virtual replica of the smoking room, capable of reading datasets from the production process and providing corresponding visualisations. The purpose is to enable automatic production management by simulating process performance when manufacturing parameters are modified, so undesirable outcomes can be prevented.

Future improvements to the current system include integrating the expert knowledge of operators in the system to validate its predictions. Sensors for weight monitoring are also being developed to monitor sausage weight loss during smoking. The integration of sensor measurements in the model would increase its precision and robustness.

These and other improvements are expected to take the digital twin to TRL9 – a commercial, modular system for the modelling and optimisation of any food processing sector.