Digital Twins for Hydroponic Lettuce Growth

Abstract

With current population trends, concerns have been raised about the capability of global food networks for coming demand, especially in high-density urban centers. Alternatives to soil-based horticulture, such as controlled environment agriculture and hydroponics, have been developed and optimized for large-scale facilities. While large facilities can supply broad markets, small-scale ventures could support specific communities and offer foods not typically produced at larger scales. Especially in schools or urban agricultural settings, automating the growth process could provide easier access to food and educational opportunities. While recent research has focused on large-scale facilities, this research provides improved automation and growth information for operators in smaller facilities.

The concept of digital twins was promising for handling biological complexities of plant growth. Modelling growth without a strict initial condition is difficult, but digital twins use a consistent stream of measured information to build a better understanding of the system. To measure growth, the measurement stream included light intensity, temperature, carbon dioxide and biomass. Initially, a dataset was collected of 1305 plant measurements, pairing RGB-D images with a mass measurement. Biomass was estimated using a custom convolutional neural network architecture that evaluated mass from an image within 1.45 g of the ground-truth. Finally, these measurements are passed to a model that defines growth, updating model parameters to reflect new information. This research evaluated possible models and chose a biological model, NiCoLet B3. After calibration, this model was able to project growth between one to four days into the future, maintaining around a 2 g forecasting error.