Fig 1.
(a): Growing configuration inside the Food Server. (b): A view inside the Food Server during experimentation.
Table 1.
Hydroponic system design elements.
Table 2.
Food Server environmental design elements.
Fig 2.
Overview of recipe optimization methodology.
First, experimenters Design Initial Recipes based on prior knowledge about the space of acceptable growing conditions. This design includes specifying the input variables and ranges that define the space of possible recipes. Second, these recipes are implemented in real-world controlled environments which Grow Plants to Maturity. Third, GC-MS is used to Measure Volatiles in mature plants. Fourth, this chemical data is aggregated to Extract Target Metric, e.g., chemscore, which is an overall indicator of flavor content. Fifth, the target metric results are used to Build Surrogate Models that model the target metric based on the input recipe variables. Sixth, a search procedure is used to Discover Optimized Recipes that are the most promising for increasing flavor according to the surrogate models. These new recipes are then implemented in the real world as the cycle repeats. The power of this method comes from the fact that modeling and optimization of flavor is done offline with automatically-built models to minimize real-world costs.
Table 3.
Treatment conditions (UV and PAR photoperiod), weight, and chemical results.
Table 4.
Spearman correlations between selected input variables and metrics.
Fig 3.
An illustration of the surrogate model and the recipes suggested by the optimization.
The three axes correspond to the three actuators and the color of the small dots indicates their value predicted by the model (i.e. flavor; red > yellow > green > blue). The large dots are suggestions, and the darker dots are the most recent ones. They suggest utilizing long photoperiods and UV periods, the success of which was confirmed in growth experiments in the Food Computer.
Fig 4.
Linear regression analysis of actual vs. calculated log R-Score for three different models.
(a): A linear model trained on UV, photoperiod, and PAR. (b): A linear model trained on photoperiod only. (c): A linear model trained on residuals after removing photoperiod effect. Photoperiod dominates the other variables (or possible there are significant nonlinear effects between these variables).
Fig 5.
The MIT expansion facility under development.
(a): Four containers being converted to large-scale Food Servers (b): The entrance to the next generation of MIT OpenAg Food Servers.