Creation of a prototype biomimetic fish to better understand impact trauma caused by hydropower turbine blade strikes

Ballistic gelatin experiments

To our knowledge, there are no published accounts of ballistic gelatin being used as a surrogate for fish tissue, so we designed several experiments to establish its baseline material properties. Ballistic gelatin was chosen because of its established use as a human and animal tissue simulant in ballistics research (Jussila, 2004; Maiden et al., 2015). Furthermore, there are well established protocols and recipes for ballistic gelatin that were easy to modify to meet our needs. We used ballistic gelatin powder specifically formulated to simulate human body density (Vyse^® Professional Grade Ballistic Gelatin; Lot #12953; Custom Collagen, Inc., Addison, Illinois, USA; customcollagen.com) for all trials and final model preparation. Our main metric to measure the material properties of gelatin was tissue durometer (i.e., material hardness or resistance to indentation) for all ballistic gelatin trials. More specifically, we measured durometer with a Shore Type-OO durometer (Model DD-4 Digital Durometer; Precision = ± 0.1 units; Rex Instruments, Buffalo Grove, Illinois, USA; durometer.com) which is best suited to measure soft gels and animal tissue. An automated stand (Model OS-1 Operating Stand, Rex Instruments, Buffalo Grove, Illinois, USA; durometer.com) lowered the meter to the sample at precisely the same rate under a consistent load pressure for all samples, thereby decreasing measurement error.

In preliminary trials, we tested two methods of ballistic gelatin preparation that were modified from other sources to accommodate our smaller sample volumes (Jussila, 2004; Cronin, 2011; Maiden et al., 2015). Method one (referred to as cooling hydration) included heating deionized water to a desired temperature using a water bath (Thermo Scientific Precision Microprocessor Controlled 280 Series Water Bath; thermofisher.com), followed by adding the heated water into a large (∼900 mL) polypropylene container containing gelatin powder. The water and gelatin were then mixed with a metal spatula until completely homogenized so that no clumps remained. At this point, up to 150 µL of de-foaming agent (Custom Collagen, Inc., Addison, Illinois, USA; customcollagen.com) was added to remove foam and excess bubbles. The mixture then cooled to room temperature (∼22 °C) which allowed the gelatin to hydrate. After this cooling hydration period, the container was covered with a lid and refrigerated for 12 h at 4 °C to allow the gelatin to completely set. Finally, the block of ballistic gelatin could be removed, cut into pieces, re-melted, and distributed as needed into other containers for testing. The second method (referred to as heated hydration) was similar to the previous except the heated water and gelatin mixture was covered with parafilm wax, placed back into the same temperature water-bath, and allowed to hydrate at this temperature for at least 10 min. Following the heated hydration period, the gelatin mixture could be distributed into test containers, allowed to cool to room temperature, and refrigerated for 12 h at 4 °C. We preferred the heated hydration method because it allowed the gelatin mixture to hydrate without cooling, avoided evaporative water loss during re-melting, and samples could be poured immediately into test containers. Both methods produced comparable estimates of durometer in our ballistic gelatin samples, but heated hydration was preferred because of more consistent heating and avoided unnecessary re-melting. Lastly, we used cinnamon oil to increase the shelf-life of our ballistic gelatin samples well beyond the normal 7 to 11-day limitation imposed on use of refrigeration alone (Jussila, 2004; Staymates & Gillen, 2010). We used cinnamon oil (NOW^® Cinnamon Cassia oil; Item #051210; gnc.com) dissolved in 95% ethanol (1:10) at a concentration of 515 ppm as a microbial growth inhibitor. Cinnamon oil was dissolved in 95% ethanol to make it more miscible in water because pure cinnamon oil extract will separate from the gelatin (Jussila, 2004). Use of the heated hydration method and cinnamon oil ensured more consistent durometer measurements during experimental trials.

Most published accounts of ballistic gelatin include use of 10 or 20% solutions (mass to volume) of gelatin powder dissolved in deionized water (Jussila, 2004; Cronin & Falzon, 2011); however, we were unsure which concentration best mimicked fish tissue. We tested a total of five concentrations including 10, 15, 20, 25, and 30% to determine which concentration best approximated the durometer of actual fish tissue (see last experiment). Each ballistic gelatin concentration was prepared in triplicate. A heated hydration protocol with a preparation temperature of 65 °C was used to create each gelatin mixture. Following hydration, ∼60 mL of each replicate was added to a 100-mL, polystyrene weigh boat and allowed to cool at room temperature. When the gelatin reached room temperature (denoted by solidification of the gelatin) all samples were labelled, placed into a large, 33 × 38 cm, 6-Mil plastic storage bag, and refrigerated overnight. Following refrigeration, three randomly selected sample weigh boats were removed and allowed to warm to room temperature for 30 min. Ten durometer measurements were recorded for each sample by removing it from the weigh boat, flipping it over, and taking measurements across the gelatin’s bottom surface. The durometer measurements for each replicate were averaged and the arithmetic mean of all three represented the average durometer of each concentration group.

There are conflicting accounts of which water temperature is best for preparation of ballistic gelatin with respect to maintaining optimal material properties of gelatin. Preparation temperatures may range from 40 to 90 °C or higher; however, manufacturers recommend temperatures near 40 °C to maintain its tissue-simulating properties (Fackler & Malinowski, 1987; Cronin & Falzon, 2011). We also experimented with preparation temperature to determine how it affected the durometer of our ballistic gelatin samples. All ballistic gelatin samples in these experiments were made using a 25% gelatin concentration. Three temperature treatments—45, 55, and 65 °C—were prepared in triplicate and durometer was measured for each sample. In addition, we also tested how warming time following refrigeration affects durometer measurements. This experiment included preparation of three replicate gelatin samples using a concentration of 25% and a water temperature of 45 °C. Following refrigeration, five durometer measurements were made immediately (time 0), every 10 min up to 1 h, then 15 min up to 2 h, and finally every 30 min for up to 4 h. Durometer was measured and reported in the same manner as the concentration experiment for each temperature and warming time treatment group.

The final set of experiments compared the material properties of ballistic gelatin to that of an actual fish to determine how closely we could mimic natural tissue. In addition to gelatin, we tested the use of an artificial skin surrogate that covered our ballistic gelatin samples. More specifically, commercially available Plasti Dip^® as a skin surrogate and found that spraying was preferable over dipping to sufficiently cover the gelatin samples. The first set of experiments was used to determine if the surrogate skin covering would significantly increase durometer compared to uncovered samples. We prepared an additional three replicates of 25% ballistic gelatin at 45 °C for use in these tests. After refrigeration, we allowed samples to warm for 30 min and took 10 durometer measurements. One layer of surrogate skin was then applied to the gelatin sample and allowed to cure for 30 min under a fume hood. The samples were then refrigerated for an additional 12 h, after which they were removed, allowed to warm for 30 min, and durometer measurements were taken. The same protocol was repeated for two, three, and four additional layers of surrogate skin for comparison. Next, we collected durometer data from three bluegill sunfish, Lepomis macrochirus, with a total length of ∼16 cm and mass of ∼90 g. All three fish were euthanized via overdose of 250 ppm clove oil in 95% ethanol (1:10) immediately prior to durometer measurements. Durometer was taken for each bluegill at 27 different locations along the entire body surface except the head which was mostly bone and the fins which were too thin to measure (Fig. 1). Another set of 27 durometer measurements were taken on the same three fish after removing scales from the entire lateral surface. The shapes of the Gelfish models and bluegill specimens required us to take durometer measurements by hand because both surfaces were curved which precluded use of the automated stand used for other ballistic gelatin experiments. In this way, we created data sets for bluegill whole-fish durometer with and without scales for comparison of ballistic gelatin with and without a skin surrogate. Average durometer was reported in the same manner as the concentration experiment for each surrogate skin layer sample, Gelfish model, and bluegill tested.

All statistical tests were performed using R v.4.0.2 statistical programing language and Sigma Plot v12. One-way analyses of variance (ANOVA) were used to compare the average durometer of (1) different ballistic gelatin concentrations prepared at 45 °C and (2) preparation temperature groups composed of 25% ballistic gelatin. A one-way repeated measures ANOVA was used to analyze the difference in average durometer between warming time and skin-layer treatment groups. Paired t-tests were used to compare average durometer between (1) Gelfish with and without Plasti Dip skin and (2) bluegill sunfish with and without scales, whereas an unpaired t-test was used to compare average durometer between Gelfish with skin and bluegill sunfish with scales. In the event a significant difference was detected by ANOVA, we used Benjamini–Hochberg post-hoc multiple comparison tests to determine the statistical relationship between treatment groups. Finally, linear regression was used to test the relationship between ballistic gelatin concentration and average durometer. All statistical decisions were based on α = 0.05.

Fish scanning and image collection

Most biometric image data used to create our Gelfish model originated from 3D scans of freshly caught fish. We scanned four species of fish including bluegill sunfish, rainbow trout (Oncorhynchus mykiss), gizzard shad (Dorosoma cepedianum), and white bass (Morone chrysops) which varied in size (Table 1). These species were chosen because they represent the range of fish body shapes that may pass through hydropower turbines and because blade strike laboratory data are available for each species (Pracheil et al., 2016a; Bevelhimer & Derolph, 2019; Bevelhimer et al., 2019; Saylor, Fortner & Bevelhimer, 2019; Saylor et al., 2020). To prepare for scanning, live fish were euthanized in an overdose of 250 ppm clove oil in 95% ethanol (1:10) for at least 15 min. Each fish was secured in an upright position with paired fins placed against the body and with the mouth and operculum closed. Individuals were frozen in this position at −20 °C for 12 h prior to scanning. Freezing was necessary to prevent movement of appendages during scanning which helped minimize image processing time. Additionally, the frozen fish was secured to a platform in an upright position that prevented movement but allowed for complete access to scan the entire fish. Finally, each fish was completely covered with a white, matte-finish spray paint to reduce surface reflections caused by fish scales. Two different scanners were used to capture fish images: a Leica Laser Tracker and Scanner (accuracy ± 0.060 mm; lieca-geosystems.com) and a FARO^® SCANARM blue light laser scanner (accuracy ± 0.075 mm; faro.com). During scanning, Verisurf software (verisurf.com/software) was used to convert images into a point cloud file.

All laser-scanned point cloud data were processed and converted into a computer-aided design (CAD) model to be used for 3D printing. The point cloud data were converted to ASCII files and imported into Geomagic Design X (3dsystems.com) software. Some noise and unneeded areas (i.e., scanning platform or fish restraint device) of the point cloud file were manually removed. The dorsal, pelvic, and anal fins were removed from the model to simplify preparation of the mold. We used internal software features like “reduce noise” with a smoothing level of 1 and default levels of “sampling” to smooth the point cloud data. After smoothing, we used the “wrap” command to transform the point cloud into a mesh. If the mesh contained more holes the images went through additional smoothing using the “fill holes” or “repair” features to close minor or larger gaps in the mesh, respectively. The final mesh was created by using “remesh”, “smooth”, and “remove spikes” features. Lastly, the final mesh was converted into a SolidWorks surface image, using the “auto surface” feature with the specifications of an organic geometry type, target patch count of 500, and default adaptive tolerance. The SolidWorks surface model was exported as a .STL file to be used in 3D printing of the fish mold.

We also investigated two additional forms of image acquisition including use of computed tomography scans of preserved specimens or from online databases. Our fifth and final species, American eel, Anguilla rostrata, was created by scanning a preserved specimen. The eel we used was much smaller than most sizes known to pass through hydropower turbines (Table 1); however, it was used to test our ability to account for and change fish size during image processing. The eel was scanned using a computed tomography scanner through the Diagnostic Imaging Service available at the University of Tennessee College of Veterinary Medicine (UTCVM). The computed tomography scan of the eel was saved as digital imaging and communications in medicine (DICOM) file. An online digital repository called Morphospace (morphospace.org) was also used to find additional X-ray, computed tomography (also computed axial tomography; CAT), or laser-scanned images for white bass. While we generated our own 3D scan data, we were also interested in how readily available online data might also be used to create fish molds. Computed tomography images (either directly imaged or taken from repositories) are not available in point cloud form, so these images were first converted into .STL files using open source InVesalius 3 (invesalius.github.io) software. A contrast range with a lower bound between 42 and 65 and an upper bound of 255 best captured skin traits and underlying skeletal structures while simultaneously filtering out noise. Next, CloudCompare (daniel.gm.net/cc/) and MeshLab (meshlab.net) or Geomagic Design X were used to convert the .STL file into a point cloud by sampling one million points, which ensured sufficient detail for the CAD model while minimizing computational resources. The conversion of point cloud to a surface model (i.e., an .STL file) followed the same procedure as that described above for laser-scanned images.

Mold printing and construction

Each CAD model was further reviewed, and final modifications were made to ensure clean demolding and purging of air during casting. The thickness of the caudal fin and peduncle was increased so that the final cast model made of ballistic gelatin would not rip when removed from the mold. The bluegill and gizzard shad CAD models only included the caudal fin, whereas the rainbow trout and white bass CAD models also included slightly raised areas on the dorsal and ventral surfaces to represent dorsal and anal fins, respectively. Additional features found in all the species CAD models were raised areas that represented the eyes and operculum on the head which were also important landmarks for positioning sensors. The eel CAD model also went through additional processing to remove its notably longer dorsal, anal, and caudal fins. Other modifications to the eel model included scaling-up body proportions to account for different sizes of eel because the original fish was smaller than most eels known to pass through turbines. The fish CAD model was then subtracted from a box to create a negative space within it, which serves as the mold for the casting. A fill hole was added to each mold CAD model on the anterior (head region) through the mouth to avoid disrupting the shape of the body and allow easy access for filling the mold with ballistic gelatin. The final mold CAD model was split in half and alignment holes, pry points, and mounting hardware were added that ensured the mold was properly sealed and aligned during casting. To additively manufacture the molds, a build file was created that contains the printer tool path and material extrusion rates. We used the Stratasys Insight software to slice the molds into layers and generate these build files, which were loaded into the Stratasys Control Center for printing. Finally, molds were printed using a Stratasys Fortus 400mc printing system and were composed of spares infill acrylonitrile butadiene styrene (ABS). The inside of each half of the final molds were polished with acetone to completely seal each surface prior to casting (Sikder et al., 2014; Lalehpour & Barari, 2016).

Sensor specification and calibration

A 3-axis digital accelerometer (ADXL375, analog.com) with a capable measurement range of ±200 g was used for data acquisition in the ballistic gelatin model during simulated blade strike impact trials. Output data was accessed through the I²C interface at the rate of 800 Hz. The I²C protocol was configured with supply voltage V_S = 3.3V, interface voltage range V_DD/IO = 3.3V and external pull-up resistors R_P = 1020 Ω (Fig. 2). The maximum pull-up resistor value (R_Pmax) was limited to 1180 Ω by the rise time (t_r) for SCL and SDA and the capacitive load on each bus line (C_b), which is given by the following equation: (1)${\displaystyle R_{Pmax}=\frac{t_r}{0.8437\times C_b}}$

Data acquisition consisted of NI cRIO 9067 (sine.ni.com), utilized as a target device and NI 9402 (sine.ni.com) module to provide the digital lines for SDA and SCL wires of I²C protocol. All hardware was programmed in LabVIEW (sine.ni.com) using the SPI and I²C Driver API, which served as the I²C master, and used the NI 9402 digital I/O to interface with the accelerometer. The LabVIEW Host code included in the API, in addition to the FPGA code, was used for initializing the accelerometer, configuring the I²C protocol parameters, data read/write, and data logging operations. Data logging frequency was set via a timed loop in the host code and stored in a .TDMS file with a local time stamp associated with each reading. Calibration of the sensor was achieved by following the single point calibration scheme specified by the original equipment manufacturer. The 0g measurements represent a potential bias in acceleration that can result in incorrect output from the sensor, so 0g measurements were specified for all three axes. This calibration scheme aligned the x- and y- axes to the 0g field, while the z-axis was oriented to the 1g field. Alignment with the 1g field also required additional sensitivity compensation of the z-axis to ensure 0g was registered correctly. All 0g offset values were then stored in the LabVIEW code and written to the dedicated offset registers during sensor initialization. The wired accelerometer was potted with black epoxy potting compound (3M-DP270, 3m.com) using a custom mold. This provided the accelerometer, and the connections with the data acquisition system, necessary mechanical rigidity and watertight seal. The potted accelerometer was embedded into the ballistic gelatin model and could be used multiple times, i.e., used in multiple ballistic models without deterioration.

Gelfish model preparation & testing

For our initial complete Gelfish model, we chose to use rainbow trout because the body depth and width of this species could better accommodate an accelerometer. During Gelfish production, the accelerometer was held in place within the rainbow trout mold using a monofilament line that stretched from head to tail. We positioned the accelerometer posterior to where the operculum would be on a real fish. This location represents the mid-body area, which is associated with the highest rates of injury and mortality when fish are struck by hydropower turbine blades, including rainbow trout (EPRI, 2008; Bevelhimer et al., 2019; Amaral et al., 2020; Saylor et al., 2020). The mold was then securely closed and kept in an upright position to cast the mold. A 25% ballistic gelatin solution was prepared at 45 °C and injected into the mold using a 60-mL syringe with an extended tip. The syringe tip was inserted completely into the mold and gelatin was injected from the bottom upwards to avoid formation of bubbles. After it was filled, the ballistic gelatin was cooled for 10 min at room temperature, followed by refrigeration at 4 °C for 90 min. Refrigeration was used to accelerate cooling and decrease the time required for the gelatin to completely set. Following refrigeration, the ballistic gelatin model was removed from the mold, placed into a sealed plastic baggie, and refrigerated again at 4 °C overnight. A surrogate skin (i.e., Plasti Dip^®) was applied after overnight refrigeration such that four separate layers were added with at least 45 min of curing time between each layer. The final Gelfish model was then placed back into the baggie and refrigerated prior to its use in blade strike impact trials.

We used the same simulated blade strike apparatus and procedure described in (Saylor et al., 2020), to strike the rainbow trout Gelfish model. We struck the Gelfish 12 times, with each strike accounting for a different velocity and leading-edge blade width, as well as a different impact location and orientation on the model itself, while all blade strikes with the model occurred at 90° (Table 2). These strike conditions were chosen based on previous laboratory tests which found that mid-body, lateral strikes caused the highest rates of injury and death among rainbow trout (Bevelhimer et al., 2019; Saylor et al., 2020), and also based on relative proximity to the embedded accelerometer. We considered impacts a “direct” sensor strike when the blade made contact with the model at the approximate center of the accelerometer. Alternatively, an “indirect” impact was considered any strike where the blade made contact with the model posterior (towards caudal fin) and away from the accelerometer (Fig. 3). All of these conditions were used to assess the ability of the accelerometer to detect differences in strike impact location on our model, which is impossible to determine on live fishes that pass through hydropower turbines. All strikes were recorded at 1000 fps with a high-speed video camera (Model IL4, Fastec Imaging, San Diego, California, USA; fastecimaging.com) and integrated stroboscope LED lighting system (Monarch Nova-Pro 300, Monarch Instrument, Amherst, New Hampshire, USA; monarchinstrument.com) for later review and to confirm blade strike impact velocity.

Data acquisition from the 3-axis accelerometer was initiated immediately prior to engaging the simulated blade strike apparatus. Estimated blade impact velocity with Gelfish was calculated using the running average of the previous 10 frames (e.g., 10 msec) prior to and including impact. Following blade strike, acceleration data were averaged over 10 ms and 30 ms intervals. Maximum acceleration (a_MAX) was determined using the following equation: (2)${\displaystyle a_{MAX}=MAX\left[\frac{1}{t_2-t_1}{\int }_{t_1}^{t_2}a\left(t\right)dt\right]}$

according to time t and the desired time interval t ₁ to t ₂ during the acceleration pulse, which is reported as acceleration of gravity (g). We estimated maximum acceleration using 10 and 30 ms running average intervals to test which interval best captured trends in acceleration. Maximum acceleration is a running average derived from National Highway Traffic Safety Administration (NHTSA) specifications for head injury criteria when using one accelerometer (Eppinger et al., 1999). Observed acceleration (α_obs) was converted to overall magnitude (across all three axes) according to the following equation: (3)${\displaystyle {\alpha }_{obs}=\sqrt{{\alpha }_x^2+{\alpha }_y^2+{\alpha }_z^2}}$

with observed values of gravitation acceleration for the x-axis (α_x), y-axis (α_y), and z-axis (α_z) at each time point, which was plotted as 10 ms and 30 ms running averages of observed acceleration against time (ms). Plots of acceleration were used to determine the relative difference in magnitude between strike impact scenarios (Table 2). In addition, we attempted to link changes in acceleration to rates of injury and mortality reported from previous blade strike impact experiments performed on live rainbow trout (Saylor et al., 2020).

Ballistic gelatin experiments

The ballistic gelatin concentration could be easily modified to account for different tissue durometers when prepared at 45  °C and a standardized durometer measurement protocol was used. In fact, average durometer °C significantly increased with every 5% increase in ballistic gelatin (prepared at 45 °C) across the entire range tested (one-way ANOVA, F_4,10 = 162.40, p < 0.001). We also detected a significant (one-way ANOVA, F_1,13 = 532.22, p < 0.0001) relationship between ballistic gelatin concentration and average durometer given by the following linear model: (4)${\displaystyle {\mathrm{D}}_{30}=1.48\times \mathrm{BG}+0.33}$

where D ₃₀ is the durometer following 30 min of warming and BG is the percentage of ballistic gelatin. Ballistic gelatin (prepared at 45 °C) concentration explained ∼97% of the variation in average durometer of the linear model (R² = 0.974; Fig. 4). Preparation temperatures of 45, 55, and 65 °C did not significantly impact the durometer of our 25% ballistic gelation samples and all three temperatures produced an average durometer of ∼35 units. Warming time significantly (one-way repeated measures ANOVA; F_14,28 = 378.96, p < 0.001) impacted average durometer for 25% ballistic gelatin prepared at 45 °C after 10 min of warming at room temperature (22.1 °C) according to Benjamini–Hochberg multiple comparison tests. Average durometer continued to decrease significantly in a linear fashion every 10 min for the first hour of warming except between the 20 to 30-min time period. The average durometer continued to decrease after each warming period but was not significant again until it warmed for 90 min. Eventually, average durometer reached its nadir near 44 units after 150 min of warming where it plateaued for the remainder of the warming experiment (Fig. 5). Ballistic gelatin temperature increased quickly to 19 °C within the first 60 min of warming and did not increase above 20 °C for the remainder of this experiment (Fig. 5).

The use of a surrogate skin increased average durometer of ballistic gelatin blanks (Table 3) and initial Gelfish models (Table 4). Addition of just one layer of surrogate skin significantly (one-way repeated measures ANOVA; F_4,8 = 323.96, p < 0.001) increased average durometer by ∼10 units, according to a Benjamini–Hochberg pairwise comparison with samples without surrogate skin. Each additional layer applied to the ballistic gelatin samples also significantly increased durometer, except between two and three layers, which were both near 57 units (Table 3). Up to four layers of surrogate skin caused average durometer to increase by nearly 20 units, to 60.2 ± 0.9 units, compared to samples without surrogate skin (average durometer = 42.3 ±0.5). Gelfish without surrogate skin (36.2 ± 0.6) had a significantly (two-tailed, dependent t-test; t = -22.209, p = 0.002) lower average durometer than the Gelfish model with surrogate skin (61.6 ± 1.4; Table 4). Similarly, bluegill sunfish with scales removed (54.0 ± 3.2) had significantly (two-tailed, dependent t-test, t = 4.9391, p = 0.039) lower average durometer than bluegill with scales intact (66.8 ± 1.8; Table 4). Lastly, average durometer of the Gelfish model with a surrogate skin was statistically indistinguishable from bluegill samples with scales according to a two-tailed, independent t-test (Fig. 6).

3D scanning and printing fish molds

We successfully laser-scanned and printed molds for four species while a fifth was successfully printed from CT scan data. Scanning frozen fish in an upright position and use of Geomagic Design X software decreased image post-processing time from nearly 40 h (manual processing) down to only 2 to 3 h (with Geomagic software). The resulting SolidWorks models contained more surface features for the rainbow trout versus the bluegill, which required markedly more processing time (Fig. 7). The SolidWorks surface models produced using this method were also easier to upload and it was easier to modify features such as fin thickness prior to printing to ensure the resulting ballistic gelatin model did not tear (Fig. 8). The time required to complete 3D printing of each mold varied by species (i.e., smaller species took less time) but was between 8 to 12 h. Printing molds upright (versus lying flat) was necessary to limit warping of the mold halves from thermal stresses and ensured the mold halves sealed completely during casing. Acetone sealing successfully prevented ballistic gelatin infiltration through the mold, which reduced cleaning and ensured consistent casting for each model. Additional mounting brackets were included on both the dorsal and ventral surface of the final mold, which allowed the accelerometer to be suspended within the ballistic gelatin using a monofilament line (Fig. 9). Multiple Gelfish were cast from the same mold and there is no indication that casting multiple models deteriorated any of the molds. Total preparation time was ∼12 h, including casting (1.5 h), model refrigeration at 4 °C (8 h), and application of four layers of surrogate skin (2.5 h) to the model prior to testing. Many Gelfish models could be created during this time if multiple molds were available.

Gelfish model testing

The Gelfish model was capable of withstanding multiple blade strike impacts at comparably high velocities (i.e., up to 11.5 m/s) without deteriorating. The rainbow trout model was successfully exposed to nine different impact scenarios before the skin separated from gelatin model; however, the accelerometer remained functional for all 12 strike tests. While the surrogate skin did separate from the gelatin during testing, the gelatin did not deteriorate and could be reused after reapplying skin layers. Similarly, the accelerometer maintained its functionality and could also be cast into another model fish. The flexibility of the model also mimicked actual rainbow trout struck under the same conditions (i.e., mid-body, lateral strikes with 52-mm blade at ∼6.8 m/s); however, overall body curvature appeared to be slightly more pronounced with the Gelfish model (Fig. 10). For example, body curvature of Gelfish was noticeably more pronounced during (+0.014s) and after (+0.024s) blade strike impact. The model also followed a similar trajectory out of the holding brackets following blade strike impact, which mimicked trials on live rainbow trout. The surrogate skin also allowed the model to maintain its integrity throughout the impact process, which ensured the entire model (head to tail) reacted similarly to real fish.

Changes in acceleration were detected in all three axes, including just prior to impact, during impact, and as the model moved away following contact with the blade (Fig. 11). Peak magnitude generally occurred 10 ms after the bow wave produced by the blade pushed the model prior to impact. The entire impact sequence took less than 30 ms to complete. The highest peak magnitude and maximum acceleration were detected from a mid-body lateral strike with a 52-mm blade moving at 11.5 m/s (Table 2; Fig. 11). All indirect strikes had noticeably lower peak magnitudes and maximum accelerations, regardless of other strike impact scenarios. Direct impacts to the mid-body ventral surface produced comparable levels of acceleration as mid-body lateral strikes and only differed in the main axis of movement caused by the strike, i.e., x-axis versus z-axis, respectively. Strikes with the same blade moving slower also had noticeably lower magnitudes—158.73 and 107.38 for the 52-mm blade moving 6.8 m/s and 76-mm blade moving at 5.0 m/s, respectively (Fig. 12). Maximum acceleration detected with a 10 ms time interval was always higher than acceleration averaged across 30 ms, regardless of group. Gelfish trials completed without surrogate skin (Trials 10 to 12) had lower values than the same trial performed on the Gelfish model with an intact surrogate skin (Trials 7 to 9; Table 2). Trends in magnitude and maximum acceleration suggest that the Gelfish model is also capable of detecting differences in impact scenarios, i.e., indirect strikes versus strikes at slower velocities or with thicker blades. A more detailed analysis of correlation with injury risk and mortality was not possible given that only one Gelfish model was tested.