As part of a current project at Cornell University, supported by the Foundation for Food and Agriculture Research (FFAR), we organized a stakeholder webinar on development of data-informed approaches to reducing food waste from primary production to consumers. The opening presentation was presented by Martin Wiedmann, the Gellert Family Professor in Food Safety, where he first covered the basics of food waste and food loss. For example, we found out that food loss and food waste are often used interchangeably, however food waste is specifically used to describe decrease in food quality and quantity due to actions and decisions made by retailers, service providers and consumers. While food loss is describing decrease in quality and quantity of food due to actions and decisions of food suppliers earlier in the chain, both food loss and food waste are greatly impacted by shelf-life of the food. Martin continued his presentation by using fluid milk as a specific example of (i) how to develop predictive models that can be used as a decision support tools to reduce waste and loss of fluid milk from grass-to-glass, and (ii) how to develop new approaches to dynamically predict product shelf-life and price.

There are two general groups of culprits responsible for spoilage of fluid milk, (i) different cold-tolerant Gram-negative bacteria, like different Pseudomonas species and coliforms, and (ii) different cold-tolerant spore-forming bacteria, like Bacillus weihenstephanensis and different Paenibacillus species. Frist group of microorganisms are generally introduced into milk as post-processing contamination while the sporeforming bacteria are typically introduced with raw milk and can survive the pasteurization process. The first presented model was developed to predict spoilage of pasteurized fluid milk due to growth of cold-tolerant sporeforming bacteria and its design followed four specific steps as part of the simulation: (i) it selects a raw bulk tank spore concentration based on concentrations that were previously determined in raw milk from different dairy farms, (ii) it selects a single sporeformer subtype that will spoil the milk based on the subtypes identified in raw milk from these dairy farms, (iii) it applies specific growth parameters for that subtype determined in laboratory growth experiments, and finally (iv) it calculates bacterial counts in milk at different days of shelf-life. The model repeats these four steps thousands of times to give us the final range of results and estimation of how confident we are in these results. A similar model was presented that predicts spoilage of pasteurized fluid milk due to post-processing contamination with cold-tolerant Gram-negative bacteria and there was also a promise of a complete model that takes into account all of the culprits responsible for spoilage of fluid milk. One glance of the future, that was presented and where these models could be used together with other digital solutions, was a container of milk equipped with a time/temperature sensor and a QR-code that would connect to a predictive model that accounts for both inherent characteristics of that specific container of milk that are related to raw milk and processing conditions used to make it as well as the time/temperature regimes it was subjected to during transport and storage. The model like this could not only give information about the shelf-life left on this specific product but also adjust the price to incentivize a purchase of older product and reduce the product waste.

The dairy predictive models (i.e. fluid milk, yogurt) as well as some produce predictive models that were presented were mostly developed based on large sets of data collected from across the state and country; however, if the data is available these models can be made specific to a single facility or a company to create something that is known as a ‘’Digital Twin’’. A ‘’Digital Twin’’ of a food processing facility captures all of the specific characteristics of that facility and as such it can function in this digital format the same way this facility functions in a real world. What this ‘’Digital Twin’’ offers to a facility or a company is a tool to test different ‘’What-If’’ scenarios, including some scenarios that would in real world require large investments or result in large losses. For example, there are number of different interventions available that can potentially be used to extend shelf-life of fluid milk. A ‘’Digital Twin’’ can help you evaluate these interventions from both cost and benefit perspective to make the most optimal decision for your product and your company.

It looks like Cornell is well on its way to develop these and other tools for the food industry. Examples presented during this webinar are relevant to the dairy and produce industry; however, the same principles can be also applied to develop tools for other commodities and other food industry needs. Identifying these needs and developing tools that will in the end prove useful to the food industry is why this type of stakeholder engagements and exchange of information are so important for the success of digital innovations in the future.

Feel free to reach out to us at at543@cornell.edu with any questions, comments, or interests for collaboration. You can also check out our new webpage dedicated to Digital Dairy where you can find all of the information on our work in this field including the recording and presentation slides from our stakeholder webinar.