3D-COAST (3-Dimensional Cultivation of Oysters Automated with Solar Tumbling)

3.2.    Research Results and Discussion

The results of this SARE funding are organized by objective, as described in the objectives and methods section.

3.2.1.     Objective 1: Construct the Next Generation 3D-COAST Prototype Raft

At the start of the project there were several improvements planned for the older 2024 model 3D-COAST raft, most of those improvements were necessary, although a few turned out to be dead-ends and were changed or abandoned as better solutions presented themselves. Most of the design work was completed as part of a MTI grant before SARE, but the actual construction of the upgrades occurred during this project. Overall, this was a very practical part of the project that was all about getting hardware ready for testing and data gathering. Just before the 2025 oyster seed arrived on the farm the third and final upgraded raft was launched. The other two rafts were retrofitted while in the water.

The primary upgrades made were: Changing the solar rotation drive system from wire to chain; improved gearbox motor and bearing design; removal of a complicated hydraulic tensioning system; and reconfiguration of basket clip system. The raft floatation modifications mentioned in the proposal did not turn out to be necessary.

The single largest issue with the 2024 3D-COAST prototype that this SARE grant helped fix was the drive system for the solar rotators. The initial design used a steel cable and pulley system, hydraulically tensioned. Figure 10 shows the cable (underlined in blue) running to pullies that would connect to the solar rotators. This cable was then tensioned hydraulically (circled in red) and all the rotators would be linked together to the drive gearbox. Overall, this system had advantages, but the cable had issues becoming tangled and slipping. In 2025 this cable was replaced with a #60 stainless steel roller chain, as shown in Figure 11. This solved the slipping issue and after some tweaking became a very reliable design by the end of the effort. Another change made to the rafts were new driveshafts running from the gearboxes (Figure 12) and new sprockets on the drive wheels for each solar rotator (Figure 13).

Another significant issue in 2024 that was fixed during this design revision was an upgraded gearbox design that used a higher quality DC motor to improve reliability related to poor brush contact, as well as a more rugged output shaft and bearing. Figure 14 shows the failed connection between the gearbox and the main driveshaft where the smaller 1” output shaft fractured the coupling dramatically due to the high torque being transmitted. On the right of Figure 14 an upgraded 2” shaft and bearing can be seen, which solved the issue permanently.

One of the smaller changes, yet highly impactful, was to modify the method of attaching the Hexcyl oyster baskets to the rungs of the solar rotators. Figure 15 shows how the baskets are clipped to the rungs using stainless steel snap hooks. In 2024 these hooks were installed on the same side of the Hexcyl baskets as the door latches, this turned out to cause major snagging and tangling. By switching which sides these hooks were installed the breakdown rate of the rafts more than halved.

3.2.2.     Objective 2: Measure Reliability and Evaluate any New Issues

For this project TWIYO developed a system for tracking the raft reliability. The metric is based on how many of the solar rotators are “down” out of the total number operating. Using this measure TWIYO can track if the system is becoming more reliable over time as design changes are made. 2025 is the first year this was carefully tracked, but there is some data from 2024 that shows the new changes made a big difference (Figure 18).

A spreadsheet that tracks every slot in the 3D-COAST rafts is maintained (a partial snippet is shown in Figure 16), one of the metrics is if the given rotator is currently broken or not. Each time the farm is visited this field is updated, allowing a tally of “breakdowns per week” to be tracked (Figure 17) and then plotted over time with a 7-day running average (Figure 18). This plot effectively shows what percentage of the raft is going down per day on average over a week. Two encouraging trends were shown: 1) Compared to 2024 the reliability of the roller chain drive and other changes cut breakdown rates by 10-20x. 2) Over the 2025 season as other improvements were made the reliability improved even more until it averaged at roughly 1% by the end of the season. 1% in this context means that each day roughly 1% of the raft capacity would need maintenance. This equates to roughly 1 out of the 48 solar rotators (2025 capacity) breaking every other day. Maintenance for one rotator was measured at approximately 20 minutes; therefore, an overall maintenance labor time can be extrapolated for different size farms. A farm using 8x 3D-COAST rafts to produce 800k-1M oysters per year could expect to spend approximately 5 hours per week on routine repairs (not including seasonal preventative maintenance). As development continues this number can be refined and tracked with additional data points to improve confidence.

During development of such systems as 3D-COAST it is expected that new issues will arise, several problems were identified during the 2025 season over the course of this project that will be addressed in 2026 and future scopes of work.

• Gearbox corrosion: Previously much of the raft structure has been made from hot-dipped galvanized steel. Over time many new stainless-steel parts were introduced into the design, often submerged near the galvanized parts. Having dissimilar metals in an electrolyte solution (salt water) caused severe galvanic corrosion of the less noble metal (the galvanized steel). This wasn’t an issue before the stainless parts were introduced, but in 2025 the quantity of stainless reached a critical point where the galvanized cases of the gearboxes began to corrode severely. The stainless parts cannot be easily replaced with galvanized steel, so the remaining submerged galvanized parts will be switched to stainless instead. This is only the gearbox cases at this time, and will add some cost to the product, but will add significant lifetime. In 2026 new gearboxes will be manufactured that will include several improvements, including new case material. The superstructure of the rafts that is above the surface of the water can remain galvanized steel and has shown good resistance to corrosion over the last 2 years of service.
• Loading solar rotators: The biggest usability issue currently facing the design is the process of loading baskets on and off each solar rotator. Currently each basket must be clipped to the rungs manually while the rotator is also manually rotated through its cycle (Figure 15). This works fairly well for someone familiar with the machine, but will present challenges for inexperienced users, and is slow. This process also involves walking out on the narrow beams that support the rotators, presenting a slip hazard. The best solution to this human-factors/usability problem has been the subject of debate for several years now. TWIYO believes that the best option to try in 2026 is the implementation of a gantry crane that can lift an entire solar rotator out of the water at once and bring it over the work platform to be unloaded. One complicating issue is that this crane will reduce stability and add enough weight that the floatation system on the raft will need to be upgraded in order to make it safe.

3.2.3.     Objective 3: Integrate the 3D-COAST Raft into Existing Farm Workflows and Measure Labor Benchmarks for Farm Efficiency and Production Modelling

The work performed under this objective produced the primary “information product” for this project by generating a useful tool which TWIYO freely supplies to the public as a way of sharing the learning from this grant. The information gathered in this section provided the data for building a sophisticated production model in MATLAB that predicts a farm’s performance given certain inputs such as number of oysters seeded, harvesting curves, growth curves, and mortality rates. These inputs are pre-populated to match 3D-COAST performance in typical mid-coast Maine waters (as measured on TWIYO’s farm), however, the model algorithms would work well for any region if a farmer knew their own parameters to calibrate the model. As far as TWIYO understands this may be one of the more sophisticated models of shellfish growth on a farm, giving detailed insight on capacity need, product availability, and labor requirements over the course of a year. As a future project TWIYO is interested in collaborating with The University of Maine to generate the growth rate curves and other input data for our model that is region specific by utilizing satellite data as described in “Using dynamic energy budget modelling and high-resolution satellite products to predict eastern oyster growth at a farm scale” by Thomas Kiffney et al[1].

The production modelling tool is hosted on TWIYO’s website (Figure 23) for free download. Figure 20 and Figure 21 show examples of some of the inputs available to a user. USDA SARE and other grantors have been acknowledged in the application as shown in Figure 22.

The following flow charts describe how the model algorithm works. Figure 24 shows the basic concept where the program loops through time by the week, each week performing all the basic farm functions: harvesting, splitting baskets, seeding oysters, and simulating new growth or death. Each of these functions vary with time, depending on the age of the oysters or seasons, for example new growth will be 0 during the winter weeks. The following charts show the details of the logic of each of the primary nodes.

The program produces a large number of charts and graphs that can be used in a financial model or for other planning purposes. Examples (actual harvest numbers are not representative of the real TWIYO farm) include Figure 29, where the total number of oysters harvested each year is predicted and broken down into size class and age groups. Other outputs include the predicted fates of specific cohorts of oysters (Figure 30), so a farmer can see what ratio of their seed will become marketable oysters vs. mortality or misshapen jumbo oysters that are hard to sell. Additional predictions include information about the dimensions of harvested oysters (Figure 31 and Figure 32) so farmers can know what to expect in terms of length/width and length/thickness ratios. One of the most useful charts is Figure 34 which shows a farmer how many baskets they will need over the years, an important metric for sizing a farm and having enough resources on hand to deal with the planning seeding and harvesting curves.

As described in the materials and methods section regarding objective 3, TWIYO seeded the latest upgraded raft with fresh oyster seed in July 2025 and took regular measurements. This data was used to design the algorithm used in the production simulation/model, as well as create the input data such as growth rate, mortality rate, and variability in oyster shape. The following plots describe this collected data and show how it was used to create various correlations in the predictive model.

For the purposes of predicting the labor costs associated with various farm tasks TWIYO logged the times it took to perform various operations on the 3D-COAST rafts. This information was used to estimate the labor and operating costs of the 3D-COAST farm for comparisons such as Table 2. The measurements are shown in Figure 36.

Examining the flow charts that describe the production modelling algorithm you can see that the conversion of basket density (oysters per liter) to oyster size (length, width, and thickness) is an important fundamental relationship to understand. This conversion is based on data collected on the TWIYO farm and is shown in Figure 37. This correlation is valid for specific growing conditions, in this case 3D-COAST, and therefore can give farmers actual information about what to expect for the dimensional shape of their oysters (as reported in Figure 31 and Figure 32). This includes important information about the quality such as cup ratio (deeper is generally considered a more desirable trait).

One of the most important pieces of data is the growth rate of the oysters. Many data points were taken as part of this project, but it remains one of the more difficult to accurately measure because several years of data is ideally needed. Over time TWIYO will be able to refine the growth curve for the 3D-COAST system and keep the inputs to the production model updated. The current dataset, thanks to this funding, provides reasonable accuracy and is validated by independent sources[2], when in doubt TWIYO has made conservative estimates that underpredict growth and performance.

Each datapoint for the growth curve is made by measuring 12-40 oysters out of a basket, recording length, width, and thickness. The number of oysters per liter is also measured and recorded. This allows the average dimensions and estimated population variance to be calculated for each sample. Figure 38 shows a snippet from the spreadsheet used to make these recordings.

This data can then be put through a pivot table to record the length of oysters vs. their age, time of year, hatchery, growing methods or other parameters, as shown in Figure 39. One of the challenges that reduced the amount of valid data for growth rate modelling was that measurements can only be taken from baskets that have never been harvested from. This is because once some of the cohort in a basket have been harvested it statistically affects the population, usually in the form of skewing it toward the slow growers (only the fastest growing oysters are harvested first, leaving the slower individuals behind). Therefore, when the growth is measured the future the rate will be represented by the slower oysters and therefore overly conservative. This can be seen in Figure 33 where the average z-scores of the oysters decrease over time as the fastest growers are removed first. Oyster length growth can be converted to the other dimensions and density via the relationships of density and dimension shown in Figure 37.

Figure 39 shows all data points available, including those from before and after the growing season really begins. You can also differentiate specific hatcheries and growing methods (static being non-3D-COAST, dynamic meaning 3D-COAST raised). In order to get a better idea of growth rates between ages during just the growing months Figure 40 was made, that crops the data to June-October and combines all points for a given age oyster. Fitting a linear line to each age-class of oyster gives 2.16 mm/week for 1-year-old oysters and 1.06 mm/week for 0-year-old oysters (seed). TWIYO believes that the 0-year-old data points are not necessarily accurate because some of the baskets got overly fouled before 3D-COAST was implemented and it skewed the rates slower than reality when properly maintained. Future year’s data will test this hypothesis. For purposes of this grant the growth rates in the production model were kept the same between age-classes, using the average of the classes at 1.6 mm/week (see Figure 41). Rather than using a linear growth curve TWIYO modified the rate to fit the shape of the growth measured in the paper by Kiffney et al[3] such that it tapered at the beginning and end of the season (Figure 42). However, on average, over a season the growth rate was made to match the nominal 1.6 mm/week measured on the 3D-COAST farm.

The last big piece of information needed as input to the production model measured during this project is the expected variance in shape of the oysters (Figure 43 and Figure 44). This is controlled in the model by setting the initial variance of the seed dimensions as well as adjusting the variance in growth rate. The measured variance and mean of L/W and L/T (Figure 43 and Figure 44) can be compared to the predicted variance and mean of L/W and L/T (Figure 32) and the model inputs adjusted in a calibration process until these measures match.

[1] (Thomas Kiffney, 2026)

[2] (Thomas Kiffney, 2026)

[3] (Thomas Kiffney, 2026)

[4] (Thomas Kiffney, 2026)