Testing the Feasibility of Selected, Commercially Available Wearable Devices in Detecting Agricultural-Related Incidents

Aaron James Etienne; William E. Field; Shawn G. Ehlers; Roger Tormoehlen; Noah Joel Haslett

doi:10.13031/jash.15985

Aaron James Etienne^1,*, William E. Field¹, Shawn G. Ehlers¹, Roger Tormoehlen¹, Noah Joel Haslett¹

¹Department of Agricultural and Biological Engineering, Purdue University, West Lafayette, Indiana, USA.

The authors have paid for open access for this article. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License https://creative commons.org/licenses/by-nc-nd/4.0/

Submitted for review on 26 February 2024 as manuscript number JASH 15985; approved for publication as a Research Article by Associate Editor Dr. Aaron Yoder and Community Editor Dr. Michael Pate of the Ergonomics, Safety, & Health Community of ASABE on 24 July 2024.

Citation: Etienne, A. J., Field, W. E., Ehlers, S. G., Tormoehlen, R., & Haslett, N. J. (2024). Testing the feasibility of selected, commercially available wearable devices in detecting agricultural-related incidents. J. Agric. Saf. Health, 30(4), 181-204. https://doi.org/10.13031/jash.15985

Abstract. A study was conducted to test a selection of commercially available wearable devices to determine their feasibility for triggering incident detection during a variety of simulated agricultural incidents with high risk of causing injury. The goal was to ultimately increase survivability outcomes for victims by enhancing notification and reducing response time from emergency services. A 50th percentile adult male anthropomorphic test device (ATD). was fitted with a convenient selection of commercially available wearable smart technologies to measure the responsiveness of the technology’s incident detection software. Devices used for this testing were: (1) Garmin Vivoactive 4 smartwatch; (2) Apple Watch Series 7 (Bluetooth only and cellular models); and (3) Movesense Active tracking device. A Samsung Galaxy S22 smartphone and an Apple iPhone 12 smartphone were used to connect the wearable devices and measured impact through their internal inertial measurement unit (IMU) sensors. Simulated ejections from equipment, vertical falls, and vehicle overturns were performed with the ATD. Side upsets were simulated with the ATD positioned in the operator station of a 52-drawbar horsepower (dbp), two-wheel drive, standard front axle, diesel tractor, weighing 6500 pounds. The tractor was equipped with an approved ROPS. Side upsets were also simulated using a 22-horsepower zero-turn mower, with the ATD positioned in the operator seat. Falls were simulated from heights of up to 4.57 meters. After each simulated incident, devices were examined to determine whether or not incident detection was successfully triggered. Data was then collected from an internal sensor logging application installed on the selected devices. It was found that the incident detection feature on the identified wearable devices only triggered in specific scenarios. Only 2 of the 27 simulated incidents successfully triggered incident detection on one device. Only the Garmin Vivoactive 4 smartwatch triggered incident detection. No device was triggered during the ATD impact in simulated tractor upset testing or in simulated zero-turn mower upset testing. It was concluded that these devices, in their current form, are not reliable for use in detecting serious agricultural-related injuries, especially considering the lack of adequate cell phone coverage in the areas in which these incidents are most likely to occur.

Keywords.Agricultural incident simulation, ATD, G-force, Incident detection, Sensor fusion, Smart watch.

Wearable devices, especially smartwatches with incident detection software, have the potential to detect severe impact to the wearer, indicative of a hard fall or crash. Once an incident has been detected by the device, an alert can be sent out to emergency services, family, coworkers, or other contacts that the wearer can specify. Based on a review of the relevant literature, wearable device incident detection has not been tested in agricultural-related scenarios likely to cause injury or fatality. The use of wearable incident detection devices in agricultural settings is currently very limited but will change as the technology becomes less cost prohibitive to be more widely adopted. Based upon a review of the literature, it was determined that one of the most significant potential applications is related to tractor and equipment rollovers, including on both ROPS- and non-ROPS-equipped tractors.

Tractor rollovers continue to be the leading cause of injury and fatality among U.S. agricultural workers (U.S. Bureau of Labor Statistics, 2020). Even with advancements in tractor technology, including advanced stability and increased use of rollover protective structures (ROPS), tractor rollover incidents account for nearly 50 percent of all farm worker related fatalities (Murphy et al., 2010; Kanoski and Bresney, 2022). There are various causes of tractor overturn fatalities. These range from intrinsic risks on an agricultural worksite, such as terrain, logistic and economic issues from investing in critical safety equipment, and long-standing convictions on general safety or confidence in the operation of machinery (Smith, 2011). Previous research on tractor rollover incidents found that greater than 99 percent of injuries and fatalities happened with tractors that are not equipped with ROPS or when the operator was not wearing a seatbelt (O'Connor, 2018). Despite this, an estimated 34 percent of the 4.7 million tractors currently in operation in the U.S. are not equipped with ROPS as of 2018 (Ayers et al., 2018). Although still significant, this is an 18 percent reduction from 2011 (Smith, 2011).

In the past, research has been conducted on developing electronic devices that would detect conditions for the risk of tractor overturning and shut off the tractor engine. An early example from over 50 years ago is shown in figure 1. During some of the first tractor tipping experiments at Purdue University, a control system was developed and implemented, which could prevent rearward tractor overturns on 0- and 11-degree slopes under “extremely unstable conditions” (Mitchell, 1969).

During testing, a magnetic tape recorder was used to measure torque, drawbar pull, clutch travel, and angular acceleration. The control system was developed using an analog computer, patched into the tractor rollover test stand using a bridge amplifier. When parameters were exceeded, the system shut down the engine.

The physics behind tractor stability are critical in determining the sequence of a tractor rollover event. When a tractor’s center of gravity is displaced beyond its base of stability, a rollover is likely to occur. This is true in both a side rollover and a rear rollover, the most common rollovers experienced (Smith, 2011; Ayers et al., 2018). A tractor will not rollover if its center of gravity remains within its base of stability. The center of gravity refers to the point of equal weight distribution on the tractor or any other machinery or equipment. This implies that 50 percent of the overall weight is distributed to the front half of the tractor and 50 percent of the overall weight is distributed to the back half of the tractor. This principle holds true from top to bottom and from side to side of the tractor. An estimated 85 percent of reported tractor rollovers in the U.S. are side rollovers (Smith, 2011; Ayers et al., 2018). The rear rollover is exceptionally dangerous due to how quickly it can occur. There is often no time, during a rear overturn, for an operator to adequately react to avoid injury or fatality. Research has shown that the critical point of no return, where the center of gravity moves over the rear axle and beyond the base of stability, can be reached in as little as 0.75 seconds (Myers and Snyder, 1996). It can take just 1.5 seconds for an incident to occur from the start of a rear rollover event.

Few studies were identified that also looked at the physical forces on tractor rollover victims (Pascuzzi, 2015; Moreschi et al., 2017). Both Italian-based studies researched different analytical methodologies to identify perceptions of risk among tractor operators, dynamics involved in serious and fatal rollover incidents, experimental simulation and incident recreation through various modeling software, and recommendations, because of testing outcomes, for changes in tractor design, utilization of equipment and devices, and changes to operator behavior. Simulated g-forces at rollover impact ranged from 22 g at the neck of the test dummy to 76 g at the thorax of the test dummy. No field or laboratory tests were performed to compare simulated results from the modeling software or to provide a means of alerting others of impact.

Although no studies were identified that used wearable devices to detect harmful incidents in agriculture, including equipment upsets, other recent studies have looked at wearable-based incident detection for falls, bicycle crashes, and impact-related incidents. Nedjai-Merrouche et al. (2021) utilized an unspecified smartphone and heart rate belt to create a fall and heart trouble detection system by fusing information from the two devices. A smartphone application was created to track detection and have the user respond if they were okay. An emergency alert message was sent out if the user did not respond within one minute. For fall detection testing, data was collected from 180 tests with three testing scenarios. Twenty volunteers performed each test scenario three times, respectively. The authors claimed 97.77% precision, 95% sensitivity, and 99.16% specificity for the fall detection service (Nedjai-Merrouche et al., 2021). Few documented studies tested device responses to actual or simulated incidents. Most identified studies reviewed the type of wearable devices used in incident detection research and the type of machine learning algorithms used to predict if an incident had occurred. For example, a study by Langer et al. (2021) proposed a conceptual smartwatch application to estimate the risk level of a given biking activity from real-time sensor feedback. Langer et al. (2021) reviewed relevant literature to identify a general risk indication metric, but no testing was undertaken and no software was created from this study.

It should be noted that the studies reviewed involving tractor rollover testing were primarily conducted on ROPS-equipped tractors. However, 21%–23% of tractors used on U.S. farms are not ROPS equipped, and the overwhelming majority of fatalities occurred while operating non-ROPS equipped tractors (Ayers et al., 2018; O’Connor, 2018; Smith, 2011). No documented studies could be identified that tested the feasibility of commercially available wearable devices with built-in incident detection software in detecting a simulated agricultural incident, including tractor rollovers.

To assess the feasibility of using the selected smart devices for agricultural incident detection, several laboratory and field tests were performed. Field testing of full-scale agricultural vehicles was used to determine the reliability, accuracy, and limitations of the selected, commercially available devices in detecting impact and occupant position during an upset.

Methods

This research aimed to assess the feasibility of the selected devices' built-in incident detection software to trigger during a vehicle upset, involving an adult male anthropomorphic test device (ATD) impact, and alert emergency contacts or Emergence Medical Services (EMS). The long-term goal was to enhance the survivability of agricultural-related incidents in which the victim was alone or was unable to initiate a request for emergency services. Incident detection was set up on the selected devices per manufacturer guidelines.

Wearable devices were selected based on the following criteria: (1) Device was commercially available (could be readily purchased from a variety of sellers) at the time of study, (2) Device had built-in incident detection software that could be set up by device user, (3) Device was less than $500 manufacturer suggested retail price (MSRP), and (4) The device’s primary function was a watch designed to be worn around the wrist. Smartphones were selected based on the following criteria: (1) One Android and one Apple operating system (OS) smartphone, to be compatible with selected smartwatches, (2) Smartphone models had to be released within the last 2 years to ensure current and future compatibility with smartwatches, and (3) Smartphones had to be commercially available at the time of study. After a market analysis of currently available smartwatches featuring incident detection was performed, a Garmin Vivoactive 4 smartwatch and an Apple Watch Series 7 smartwatch were selected for the study. Apple Watch Series 7 (one Bluetooth only and one cellular-capable) smartwatches were selected, to test whether the cellular watch could detect and notify incidents without connection to a smartphone. An Apple iPhone 12 was selected to connect to the Apple Watch Series 7 smartwatches. A Samsung Galaxy S22 smartphone was chosen to connect to the Garmin Vivoactive 4 smartwatch.

Devices examined for potential use in this study but not ultimately selected included the Samsung Galaxy Watch, Garmin Forerunner, Garmin Fenix, Garmin Venu, Huawei Watch, and Google Pixel Watch. The Samsung Galaxy Watch 5 and Apple Watch Series 8 and 9 were not yet available at the time of device selection. For smartphones, the Apple iPhone 10 and 11, Samsung Galaxy S21 and S20, Google Pixel, and Motorola Edge were considered. The iPhone 13 and 14 were not yet available at the time of device selection.

Of the devices analyzed, only the Garmin Vivoactive 4 smartwatch and Apple Series 7 (both Bluetooth and cellular) smartwatches featured incident detection software. Apple iPhone 12 and Samsung Galaxy S22 smartphones were used to connect the smartwatches to the sensor logging application. Smartwatches also required connection to the smartphones for incident detection to be active and send alerts, even on the cellular Apple Watch. IMU sensor data from the Apple iPhone 12 and Samsung Galaxy S22 smartphones, as well as the Movesense Active tracking device, were recorded and analyzed after each experimental replicate in which they were tested. These devices were used to compare IMU data logged from the smartwatches and for analysis as to why incident detection was successful or unsuccessful.

The accelerometer sensors in the tested smartphones, smartwatches, and tracking devices take inertial measurements of velocity and position. The accelerometer takes measurements along three orthogonal axes: X, Y, and Z. The X axis is parallel with the device’s screen or front face and aligned with the top and bottom edges in a left-to-right direction. The Y axis is parallel with the device’s screen or front face and aligned with the left and right edges in a top-to-bottom direction. The Z axis faces perpendicular to the device’s screen or front face, pointing in the up direction (Lashkari, 2019). Acceleration from the accelerometer sensor is measured in meters per second squared (m/s²). Acceleration is converted from m/s² to gravitational force (g-force or g) to relate the ATD’s impact to earth’s gravity. One ‘g’ is equivalent to 9.81 m/s² of acceleration. For example, there is a sudden increase in acceleration from when a tractor starts to tip over to when it impacts the ground, as demonstrated by Lashkari (2019).

The gyroscope sensors in the tested smartphones and smartwatches measure the orientation and rotation of the device. The gyroscope also measures along three orthogonal axes, aligned in the same manner as the accelerometer. Angular velocity from the gyroscope sensor is measured in degrees per second (deg/s). Angular velocity is converted from deg/s to rotations per second through dividing by 360 degrees. This is done to relate the number of times a sensor is measuring rotation during the duration of an experiment. If a device’s gyroscope records 3 rotations per second over a 30-second duration, 90 rotations of the device were measured. However, for the experiments conducted in this study, there were spikes in gyroscope data during the duration of the simulated incident. In a tractor upset, for example, it might only take a couple of seconds from when the tractor is tipped to when impact with the ground occurs. Over this period, a device’s gyroscope has the capacity to measure a range of rotation from .1 to 2 rotations per second. These can be summed to find out how many times the device rotated during the upset. The equation for velocity of a free-falling object is given below.

The magnetometer sensor measures the magnetic field for the X, Y, and Z axes of the device in the same manner as the accelerometer and gyroscope sensors. The magnetometer is measured in the Gaussian unit, gauss (G), which measures magnetic induction.

When incident detection is triggered on the Apple Watch Series 7 smartwatches, a beeping and vibrating alarm is triggered to signal that a hard fall has taken place. The user can select and press the “I’m OK” button on the Apple Watch’s screen to cancel EMS and emergency contact notification of the fall. Otherwise, the alert will simultaneously be sent out to the emergency contact and EMS after 30 seconds if there is access to cellular coverage. EMS and emergency contact notification can also be disabled in the Apple Watch and iPhone settings. When incident detection is triggered on the Garmin Vivoactive 4 smartwatch, a notification alert sound is triggered, emanating from the watch. A text message is immediately sent to the connected phone (a notification alert sound will also emanate from the phone), and an email will immediately be sent to the email address identified in the Garmin account. An example of incident detection and emergency response is visualized in figure 2. As of 2023, Apple IOS 13, as well as the relevant Android Operating System (OS) version (on applicable smartphone devices that feature incident detection), will connect to a satellite communication system for emergency notifications (Apple Inc., 2023b).

Analysis of the selected wearable device sensor data, collected during Experiments 1–6, included graphing of IMU (accelerometer, gyroscope, and magnetometer) sensor data, reasoning as to what caused noise, if any, in the sensor data, and comparison between sensor data of tested devices with regards to their placement and movement during testing. Sensor data charts were created in Microsoft Excel as well as a Python script. Data was exported from the sensor logging application on the tested devices (Choi, 2023) to a ZIP file containing the sensor logs. Sensor logs were then converted from comma-separated value (CSV) files to Microsoft Excel workbooks or into a data frame in Python. In addition, sensor fusion was performed on the IMU data from each tested device. Euler angles (roll, pitch, and yaw), or the successive planar rotation angles around the X, Y, and Z axes (Ohkami, 2003), were calculated with IMU data from each simulated incident, using a Kalman filter. A Kalman filter estimates measured input variable values over a set period and accounts for uncertainty within the reviewed measurements. It also accounts for relations between estimated input variables (Tomer, 2022). G-force (g) is a measure of the acceleration produced by Earth’s gravity onto an object (Grupo One Air, 2022).

Test scenarios were determined based on the findings from A Summary of Lone Agricultural Worker Injuries and Fatalities (Etienne et al., 2023), as well as trends in agricultural incident occurrence and severity from other agricultural injury and fatality reporting sources (Nour et al., 2021). Tractor rollover/runover, slips and falls, and UTV/ATV rollover/runover were found to be among the leading causes of agricultural incidents, especially pertaining to lone agricultural worker incidents.

Representatives from Garmin were consulted before, during, and after experimentation. However, no specifics were provided on how Garmin tested their devices. Garmin devices were subject to the same impacts as other devices tested. From discussion with Garmin representatives, it was found that a physical exercise activity of the user had to be active for incident detection to trigger. For example, a ‘walk’ activity had to be selected and active, for incident detection to trigger in the event of an impact. Several attempts to contact Apple concerning their devices were made, but no response was given. All information on setup and activation of incident detection on the Apple smartwatches came from an official Apple website (Apple Inc., 2023b). Although not specified, exercise activities were also tested on the Apple smartwatches to see if activating an activity was more likely to trigger incident detection.

Methods for Incident Detection Software Setup

For the Garmin Vivoactive 4 smartwatch, the Garmin Connect application on the Samsung Galaxy S22 was first opened (Garmin, 2023). On the options menu of the application, ‘Safety and Tracking’ was selected. The ‘Safety Features’ option was then selected. Requirements for incident detection use had to first be accepted before continuing setup. The ‘My Information’ tab was then selected. Name, city, and state were added on this page. The ‘Add Emergency Contact’ tab was then selected. The newly created contact was then selected, and a request was sent to that contact. Incident detection was then enabled. For incident detection to be active on this device, an activity profile had to be started. For example, a walking activity profile was selected from the ‘Exercises’ list on the Garmin watch before testing was performed.

For the Apple Series 7 Bluetooth and cellular smartwatches, the Watch application was opened on the Apple iPhone 12 smartphone, and the ‘Emergency SOS’ button was tapped. Fall detection was turned on, and the ‘Always’ option was selected from the ‘Fall Detection’ options menu. If the Apple Watch user entered their age as above 55 when initially setting up their watch, or in the Health application on their iPhone, the fall detection feature was automatically turned on. The fall detection feature only works for users 18 and older (or if an age greater than or equal to 18 is selected when setting up the Apple Health Profile) (Apple Inc., 2023c).

To set up medical identification and add emergency contacts, the ‘Settings’ application on the iPhone was opened, ‘Health’ was tapped, and ‘Medical ID’ was selected. Date of birth and pertinent health information was then added. To add emergency contacts, the ‘plus’ button was pressed under ‘Emergency Contacts’. The contact must have been previously set up in ‘Contacts’. A contact was selected, and a relationship to the user was added. To make the medical ID viewable from the lock screen of the iPhone or Apple Watch, the ‘Show When Locked’ option was turned on. To share medical ID with first responders, the ‘Share During Emergency Call’ was turned on. Finally, ‘Done’ was tapped to finish set up (Apple Inc., 2023c).

Methods for Experimental Setup

The five experiments undertaken in this study are summarized in table 1, including the experiment date, a brief description of the experiment (with devices used during that experiment), the number of replicates in each experiment, and the total number of simulated incidents in each experiment (note: the second half of Experiment 2 was performed at a later date). For each experiment, all devices specified were tested for the number of replicates specified.

Experiment 1 was performed with a 50^th percentile ATD and a John Deere Pro Gator utility terrain vehicle (UTV). A Garmin Vivoactive 4 (Bluetooth) smartwatch was affixed to the left wrist of the ATD, while an Apple Watch Series 7 (Bluetooth) smartwatch was affixed to the right wrist. The ATD was fitted with overalls, shown in figure 3. Both an Apple iPhone 12 (left side) and a Samsung Galaxy S22 (right side) were inserted in the center chest pocket with screens facing out.

Prior to beginning the test, a sensor logger application (Sensors Toolbox—Multitool) (Apple Inc., 2023a) was downloaded on both smartphones. As this version of the application did not have a record feature, the application was left open, screen time-out was disabled, and screen recording was used on each phone. Four replications of Experiment 1 were performed, during which the ATD was physically pushed out of the passenger seat of the UTV onto the ground while the UTV was in motion. During Experiment 1, Replicate 1, the ATD was ejected onto a gravel surface while the UTV was traveling at 9.66 kilometers per hour (KPH) (6 miles per hour (MPH). After ejection, the ATD was left on the ground for one minute. This is shown in figure 4. After this period, devices were checked to examine whether incident detection was triggered. The screen record was also stopped on each phone, and an initial check was done to ensure data was collected without error. Replicate 2 was performed, with the UTV speed increased to 12.87 KPH (8 MPH). The same protocol was followed from Replicate 1, for setup, execution, and analysis. A similar impact to Replicate 1 occurred.

In Experiment 1, Replicate 3, another variation of a fall was performed by pushing the ATD out of the bed of the UTV while in motion. The tailgate of the UTV was opened and lowered. The ATD was positioned flat in the UTV bed, face up, with feet near the back of the bed. Screen recording was started on the phones, and the sensor logging apps were activated. The UTV was then accelerated to 9.66 KPH (6 MPH), and the ATD was shoved out of the back of the bed onto a gravel surface. After one minute, screen recording on the phones was stopped and devices were checked to see if incident detection had been triggered. In Replicate 4, the speed of the UTV was increased to 12.87 KPH (8 MPH). The same protocol was followed from Replicate 3.

In Experiment 2, a vertical drop test was performed to represent a free fall from equipment or structure by dropping the ATD from the forks of a forklift at varying heights. Eight replications were performed for this experiment. A Hyster battery-electric forklift was used to simulate falls at 1.52 meters (m) (5 feet (ft)), 2.13 m (7 ft), 2.74 m (9 ft), 3.66 m (12 ft), and 4.57 m (15 ft). In Replicates 1–4 of Experiment 2, forklift forks were first raised to the selected height. The ATD was then physically pushed off of the forks and onto the concrete surface. In Replicates 4–8, the forks were tilted downward once they were at the selected height, and the ATD fell onto a gravel and grass surface, respectively. Experimental set up is shown in figure 5. For Experiment 2, a new sensor logging application was used (Choi, 2023) on the Apple iPhone 12, Apple Watch Series 7 Bluetooth, and the Samsung Galaxy S22 smartphone. This application was chosen for its live record feature and ability to log sensor data on the Apple Watch. Although the sensor logging application can run on the Apple Watch, it cannot run asynchronously of the Apple iPhone application. Therefore, only one watch or the iPhone’s sensors can be logged at a time.

For Replicates 1–4 of Experiment 2, the Apple Bluetooth smartwatch and iPhone were utilized, as well as the Garmin smartwatch and Samsung Galaxy smartphone. For Replicates 4–8, the Apple cellular smartwatch and Movesense Active tracking device were added for testing. The Apple Watch Series 7 Bluetooth was logged from the Sensor Logger application on a separate iPhone 14 Pro. This phone was not placed on the ATD but was within range for an active Bluetooth connection. This was done so both Apple smartwatches could be logged at the same time.

In Replicates 1–4 of Experiment 2, the Apple Watch Bluetooth was secured to the left wrist of the ATD. The Garmin watch was fastened to the right wrist. Smartphones were placed in the chest pocket of the overalls, with the screens facing out. The iPhone was placed on the left side of the pocket, and the Galaxy phone was placed on the right side. The ATD was placed horizontally, in the middle of the forks. Once the forks were raised to the selected height, the sensor logging application was started on the Galaxy phone and for the Apple Watch Bluetooth. The ATD was then pushed off the forks. After the ATD impacted the ground, one minute was taken before stopping the sensor logger application and checking for incident detection triggering.

In Experiment 2, Replicate 1, the ATD was ejected from the forks at a height of 142.24 cm (56 in). Sensor logging was activated on the Apple Bluetooth watch and Galaxy phone. During Experiment 2, Replicate 2, fork height was increased to 198.12 cm (78 in). In Replicate 3, fork height was increased to 243.84 cm (96 in). For Replicate 4, fork height remained the same as in Replicate 3, but sensor logging was changed from the Apple Watch Bluetooth to the iPhone’s internal sensors.

In Experiment 2, Replicates 5 and 6, a drop was performed from a height of 3.66 m onto a gravel surface. In Replicates 7 and 8, a drop was performed from a height of 4.57 m onto a grass surface. IMU sensor logging for Experiment 2, Replicates 5–8, was performed on the Apple and Garmin watches, as well as the Samsung Galaxy S22 smartphone, using the Sensor Logger application. IMU sensor data from the Movesense active tracking device was also logged using the Movesense Showcase application. The Apple Watch Series 7 Bluetooth and Movesense sensors were affixed to the right wrist of the ATD. The Apple iPhone 12 and Samsung Galaxy S22 smartphones were placed in the right and left waist pockets of the ATD’s coveralls, respectively. The Apple cellular and Garmin watches were secured to the left wrist of the ATD.

In Experiment 3, the Garmin Vivoactive 4 smartwatch, Apple Series 7 cellular smartwatch, Apple iPhone 12, and Samsung Galaxy S22 devices were tested to see if incident detection could be triggered from a simulated fall out of a skid steer loader bucket. In this experiment, the ATD was dropped vertically from the bucket of a John Deere 324 G skid steer loader. The Galaxy smartphone was placed in the left side of the overall pocket of the ATD, with the screen facing inwards. The iPhone was placed on the right side of the overall pocket of the ATD, with the screen facing inward. The Garmin watch was fastened to the left wrist of the ATD, while the Apple Watch was fastened to its right wrist. Four replications were performed for this experiment. Before each replication, the sensor logging application was started on the Apple Watch and Galaxy phone. After each fall, the ATD remained static on the ground for one minute before the logger was stopped and devices were checked for incident detection triggering. Experiment 3 setup is shown in figure 6.

In Experiment 3, Replicate 1, the skid steer bucket was raised to 243.84 cm (96 in). The bucket was then turned downward using the actuator controller arm in the cab. The ATD then fell vertically out of the bucket and onto the flat concrete surface below. In Experiment 3, Replicate 2, the skid steer bucket was raised to 259.33 cm (102.1 in), the maximum height that the bucket arm could reach (Deere and Company, 2023). For Experiment 3, Replicate 3, the ATD was again dropped from a height of 259.33 cm, but onto a gravel surface.

In Experiment 3, Replicate 4, the skid steer bucket with ATD inside was raised to 50 percent of its maximum height (129.67 cm) (51 in), and the skid steer was driven around in a figure 8 pattern for a total of two minutes. The skid steer then came to a stop on the concrete pad, and its arm was raised to maximum height. The ATD was then immediately dropped out of the bucket. In this way, there was constant movement acting on the ATD and tested devices. Experiment 3, Replicate 4, was then repeated. While the forklift drop test in Experiment 2 simulated a fall onto the rear side of the ATD, the skid steer drop simulated a fall and impact onto the front side of the ATD.

In Experiment 4, 6 tractor side upsets were performed onto a gravel surface. Side upsets were simulated with the ATD positioned in the operator station of a 56-hp, two-wheel drive, standard front axle, diesel tractor weighing 6,500 lbs. The tractor was equipped with an approved ROPS with the cab glass removed. Upsets were conducted in which the tractor was rolled onto its right and left sides, three times each. A Yale 50VX forklift, with an attached Pintle hitch (16,000 lb. capacity), was utilized to induce the side upsets. Between replicates of Experiment 4, sensor recording was stopped on all tested devices, and smartwatches were checked to see if incident detection had been triggered.

A right-side tractor upset was simulated in Experiment 4, Replicate 1. The Garmin smartwatch was fastened to the ATD’s right wrist, while the Apple cellular watch was fastened to its left wrist. A Movesense Active tracking device sensor was attached to a chest strap and fastened below the ATD’s chest, above its clothing. It was used to compare readings from the other selected wearable devices tested in Experiment 4. Accelerometer, gyroscope, and magnetometer readings were selected to be measured with the Movesense Active sensor. The Samsung Galaxy S22 smartphone was inserted into the chest pocket of the ATD’s overalls, with screen facing in. The Apple iPhone 12 smartphone was placed on the dash of the tractor, with the screen facing up. Before a tractor upset was performed, an exercise activity was started on the Garmin smartwatch. The experimental setup for Experiment 4 is shown in figure 7. A strap was used to secure the ATD in the operator station of the tractor.

In Experiment 4, Replicate 2, a left upset was performed. A left-side upset was performed to compare device impact and ATD movement to a right-side upset. This was also done to see if the impact was greater on the devices secured to the left wrist during a left-side upset. The iPhone and Galaxy smartphone placements were switched from pocket to dash and vice versa. The exercise activity was kept active on the Garmin smartwatch. All other experimental factors remained the same. Left-side upsets were also performed for Experiment 4, Replicates 3 and 4, with all other experimental factors remaining the same. In Experiment 4, Replicate 5, a right upset was performed. The Movesense active sensor was moved from the ATD’s chest to its right wrist. The Galaxy phone was again moved to the tractor’s dashboard, and the iPhone was placed in the ATD’s overall pocket, with the screen facing inward. All other experimental factors remained the same. In Experiment 4, Replicate 6, a right-side upset was again performed. All experimental factors remained the same as the previous replication.

Experiment 5 simulated left- and right-side upsets with a gasoline-powered, 22-hp zero-turn mower. Five replications were performed for this experiment. The ATD was placed in the operator station of the mower. Setup for Experiment 5 is shown in figure 8.

During Experiment 5, the Apple Bluetooth-only watch and Garmin watch were fastened to the ATD’s left wrist. The Apple cellular watch and Movesense Active tracking device were fastened to the right wrist of the ATD. IMU data was collected with the Apple and Garmin watches, Movesense Active, and Galaxy smartphones. Data was not collected with the Apple iPhone, as the Apple cellular watch was being recorded from the Sensor Logger application. The Apple Watch Series 7 Bluetooth was logged from the Sensor Logging application on a separate iPhone 14 Pro. This phone was not placed on the ATD or zero-turn mower but was within range for an active Bluetooth connection. For Replicates 1 and 2 of Experiment 5, the Apple iPhone 12 was placed in the right waist pocket of the ATD’s coveralls, and the Samsung Galaxy S22 smartphone was placed in the left waist pocket, both with screens facing in. For Replicates 3–5 of Experiment 5, the phones were moved to the chest pocket of the ATD, with screens facing in. Phone placement was changed to see if IMU sensor data on the Samsung Galaxy S22 smartphone would vary.

In Experiment 5, Replicate 1, the zero-turn mower was placed at the side of a hill with a 30-degree slope. A right-side upset was induced using the forks of a skid steer loader. The ATD was fastened with a strap to the operator station of the mower to simulate wearing a seatbelt. In Replicate 2, the mower was dropped 1.5 m (5 ft) onto its right side from the skid steer forks. This was done to simulate the velocity from an upset if the mower had been moving at a speed of 19.5 KPH (12 MPH) at the time of impact. Velocity was determined from the gravitational acceleration of the mower in free fall and the amount of time it took the mower to impact the ground once dropped. It took the mower 0.55 seconds to impact the ground.

In Experiment 5, Replicates 3–5, the ATD was not fastened to the operator station. In Experiment 5, Replicate 3, a left-side upset was simulated in the same manner as in Experiment 5, Replicate 2. In Experiment 5, Replicate 4, a left-side upset was simulated with the mower on the ground. The upset was performed on flat, grass terrain, using the skid steer forks to initiate the upset. This was done to compare sensor data from Experiment 5, Replicate 1. In Experiment 5, Replicate 5, the mower was dropped 3 m (10 ft) onto its right side from the skid steer forks. This was done to simulate an upset if the mower had been moving at a speed of 28 KPH (17 MPH) at the time of impact. Although 28 KPH is likely faster than the mower will be traveling during operation, this velocity was tested to see if an extreme scenario would trigger incident detection.

Results

A total of five experiments were conducted to determine the feasibility of triggering incident detection on a selection of wearable devices in simulated agricultural incident scenarios. In four experiments, a 50^th percentile adult male (ATD) was fitted with a Garmin Vivoactive 4 smartwatch and Apple Watch Series 7 (one Bluetooth only and one cellular-capable) smartwatches to measure the responsiveness of the technology’s incident detection software at triggering during simulated ejections, vertical falls, and vehicle overturns. In addition, the IMU sensor data of the smartwatches was logged during experimentation, exported, and compared to the IMU sensor data of the other selected devices. Samsung Galaxy S22 and Apple iPhone 12 smartphones were used to connect the wearable devices, as well as measure impact through their IMU sensors. A Movesense Active tracking device was fitted to the ATD in Experiments 2, 4, and 5 to compare IMU sensor data to the tested smartwatches and smartphones. It was found that incident detection on the tested smartwatches only triggered in specific scenarios: when an exercise activity was started and constant movement occurred during a simulated incident (Garmin watch).

Only drop testing of the ATD in Experiment 3 successfully triggered incident detection. Only one experiment successfully triggered incident detection. This only occurred in 2 out of 4 replications (50%). Only 2 of 27 total replications (7.4%) performed were successful. Only 2 of the 27 total simulated incident impacts performed (7.4%) successfully triggered incident detection. Sensor readings from tested devices were collected and analyzed after each experiment. Only the Garmin Vivoactive 4 smartwatch was able to successfully trigger incident detection without a biometric signature. No experimental replication was able to succinctly trigger incident detection. UTV ejection testing in Experiment 1, forklift drop testing in Experiment 2, tractor rollover testing in Experiment 4, and zero-turn mower rollover testing in Experiment 5 failed to trigger incident detection.

During Experiment 1, Replicates 1 and 2, incident detection failed to trigger on either of the tested smartwatches, as the ATD was pushed out of the UTV passenger seat and impacted the gravel surface. In Experiment 1, Replicates 3 and 4, incident detection was not triggered on either of the tested smartwatches, as the ATD was pushed out of the UTV bed and impacted the gravel surface. Although incident detection failed to trigger on the tested smartwatches, IMU sensor data was logged during each Replicate, on the iPhone 12 and Galaxy S22 smartphones. The sensor logging application used for this experiment was not capable of logging the tested smartwatches. Sensor readings from Experiment 1, Replicate 2, taken from the Apple iPhone 12 are shown in figure 9.

Maximum g-force detected during Experiment 1 occurred in Replicate 2, with 6.05 g occurring on impact. This reading was taken from the Apple iPhone 12.

Experiment 2, Replicates 1–4, failed to trigger incident detection on the tested smartwatches during ground impact of the ATD. IMU sensor readings and resulting Euler angles on the Galaxy smartphone, from Experiment 2, Replication 3, are shown in figure 10.

Maximum g-force detected during Experiment 2, Replicates 1–4 occurred in Replicate 3, with 10.26 g occurring on impact. This reading was taken from the Samsung Galaxy S22.

Experiment 2, Replicates 5–8, also failed to trigger incident detection on the tested smartwatches. In Experiment 2, Replicates 5 and 6, IMU sensor data was collected from the Apple Cellular Watch, Movesense tracker, and Galaxy phone. In Experiment 2, Replicates 7 and 8, IMU data from the Apple Bluetooth watch was also logged. Results of Experiment 2, Replicate 8, are shown in figure 11.

Noise was present in the acceleration and gyroscope data shown in figure 11. This was due to the movement of the forklift forks. Noise was most present when shaking the forks back and forth to shake the ATD off the forks and induce the fall. The forks did not tilt down at a steep enough angle to slide the ATD off in one motion. It was found that the force of impact between the ATD and the ground surface remained similar over concrete, gravel, and grass surfaces. Maximum g-force detected during Experiment 2, Replicates 5–8, occurred in Replicate 8, with 13.69 g occurring on impact. This reading was taken from the Apple Watch Series 7 Cellular.

For Experiment 3, the Apple Watch Series 7 Cellular smartwatch used in testing was not logged with the sensor logging application. The exercise activity had to be active on the Apple smartwatch, and the sensor logging application could not be used simultaneously. IMU sensor data was instead logged on the Apple iPhone 12 smartphone, in addition to the Samsung Galaxy S22 smartphone. In Replicates 1–3, incident detection failed to trigger on either of the tested smartwatches, as the ATD fell from the skid steer bucket and impacted the concrete pad or gravel surface. In Replicate 4, incident detection was immediately triggered, and a notification alert sound emanated from the Garmin watch as the ATD impacted the concrete pad. A text message was immediately sent to the Galaxy phone (a notification alert sound was also given from the phone), and an alert was immediately sent to the email set up in the Garmin account. A successful incident detection trigger on the Garmin watch is shown in figure 12. The Apple cellular watch was unable to successfully trigger during Replicate 4.

Experiment 3, Replicate 3, was repeated, again successfully triggering incident detection on the Garmin watch. Apple iPhone 12 IMU sensor and derived Euler angle data during Experiment 3, Replicate 3, are shown in figure 13.

The vibrations and movement of the ATD, while being moved in the skid steer bucket, account for the noise shown in the accelerometer and gyroscope sensors in figure 13. The maximum g-force detected during Experiment 3 occurred in Replicate 4, with 6.42 g occurring on impact. This reading was taken from the Apple iPhone 12. Experiment 3, Replicate 4 was repeated five more times without successfully triggering incident detection. During Experiment 3, Replicate 5, incident detection was successfully triggered. As in Experiment 3, Replicate 4, the sensor log was analyzed, and meta data was recorded, to give more information on the activity and outcome. Experiment 3, Replication 5, was repeated five more times, without further success in triggering incident detection. Lastly, Experiment 3, Replicate 6, failed to trigger incident detection. Maximum g-force detected during Experiment 3, Replicates 1–5, occurred in Replicate 5, with 1.25 g occurring on impact. This reading was taken from the Apple Watch Series 7 Bluetooth.

All Replicates in Experiment 4 failed to trigger incident detection. In Experiment 4, Replicate 1, IMU data was collected from the internal sensors on the Galaxy phone, Apple Watch cellular, and the Movesense Active sensor. This was done for all subsequent replications (Replicates 2–6) of Experiment 5. IMU sensor data and derived Euler angles from Experiment 4, Replicate 2 is shown in figure 14.

Maximum g-force detected during Experiment 4 occurred in Replicate 6, with 9.44 g occurring on impact. This reading was taken from the Movesense Active.

The tested smartwatches failed to detect a zero-turn mower rollover in Experiment 5. All Replicates of Experiment 5 failed to trigger incident detection. For Replicates 1–5 of Experiment 5, IMU sensor data was collected from Apple cellular and Bluetooth-only smartwatches, a Garmin smartwatch, a Movesense Active tracking device, and a Samsung Galaxy S22 smartphone. Results from Experiment 5, Replicate 4, are shown in figure 15.

Noise in the IMU sensor data shown in figure 15 was attributed to lifting the mower into position with the skid steer forks. Maximum g-force detected during Experiment 5 occurred in Replicate 5, with 8.75 g occurring on impact. This reading was taken from the Movesense Active.

Figure 11. Experiment 2, Replicate 8, IMU sensor and Euler angle data for Movesense Active tracking device.

Figure 12. Successful triggering of incident detection on Garmin Vivoactive 4 smartwatch.

Figure 13. IMU sensor and Euler angle data from the Apple iPhone 12 in Experiment 3, Replicate 4.

Figure 14. Experiment 4, Replicate 2, IMU sensor data for Apple Watch Series 7 Cellular smartwatch.

Conclusions

This study aimed to evaluate the effectiveness of commercially available wearable devices in detecting agricultural-related incidents, specifically simulated ejections, falls, and upsets using an anthropomorphic test device (ATD). Despite the promising potential of these devices, the results were inconclusive. Only two out of twenty-seven simulated incidents triggered detection on the tested wearable devices, highlighting a significant gap in the current technology's ability to accurately identify agricultural incidents. This indicates a critical need for further research and development in this area to improve the detection capabilities of wearable devices in agricultural settings.

The findings of this study underscore the importance of collaboration between researchers and manufacturers of wearable incident detection devices. Such partnerships are essential to clearly identifying the potential applications and limitations of these devices in agricultural environments. By refining the fall detection algorithms and improving the testing procedures, it is possible to enhance the reliability and accuracy of these devices. This could lead to significant advancements in the integration of wearable technologies into agricultural safety measures, ultimately improving the survivability outcomes for agricultural workers.

Future research should focus on developing more robust testing procedures and exploring innovative algorithm enhancements tailored specifically for agricultural use cases. Additionally, there is a need to address the challenges posed by inadequate cell phone coverage in rural areas where these incidents are most likely to occur. By overcoming these obstacles, wearable technologies can become a vital component of agricultural safety protocols, providing timely alerts and reducing response times in emergency situations. The continued evolution of these technologies holds great promise for enhancing the safety and well-being of agricultural workers.

Limitations

The convenient selection of commercially available, wearable devices used for this study was not a broad representation of all commercially available, wearable devices featuring an incident detection feature. More research should be conducted on other currently available devices and devices that come to market in the future.

For incident detection to trigger on the Garmin Vivoactive 4 smartwatch, an exercise activity had to be active (Garmin, 2023). This is not feasible for normal agronomic activities, especially while operating agricultural equipment. There is currently no way to activate an incident detection alert with a currently available Garmin smartwatch during day-to-day use, without activating an exercise activity. While Apple claimed that the user can set incident detection to “always active” on the Apple Watch Series 7 Bluetooth and cellular versions (incident detection is not always active unless the detection setting is manually changed), this study could not successfully trigger detection without a biometric signature. It can be assumed that while performing an agronomic task, a person will be wearing the device and thus produce a biometric signal. It was not possible to test UTV ejection, forklift and skid steer drops, or tractor and zero-turn mower upsets with a human test subject. It was also not possible to replicate a biometric signature on the ATD used in these experiments. Future studies will look at newer smartwatch and smartphone devices with crash detection and more advanced incident detection features.

During simulated tractor side upset testing in Experiment 4, neither of the detection-capable devices (Apple Watch Series 7 Cellular or Garmin Vivoactive 4) were able to successfully trigger incident detection. Upon analysis of the IMU sensor data from the Apple Watch Cellular, it was clear that a significant impact was detected by the sensors during each replication, even though an incident was not detected. The impact imposed on the Apple Watch Cellular’s accelerometer sensor during every replicate of Experiment 4 was greater than the impact that triggered incident detection on the Garmin Vivoactive watch during Experiment 3, Replicate 3 (9.44 g at maximum vs. 6.42 g). It is likely that the Apple Watch Cellular required a biometric signature (being attached to a living person) for incident detection to be active, though no literature could be found that expressly stated this. Another possible limitation during Experiment 4 was that the ATD was strapped into the operator station of the tractor. If the ATD could move freely throughout the cab, it is possible that inertial forces on the ATD would be greater during side upset impacts.

The same issue occurred during Experiments 2 and 5. In Experiment 2, the Apple Watch Series 7 Cellular measured a maximum impact of 13.69 g. This is greater than twice the g-force from which the Garmin Vivoactive watch triggered incident detection in Experiment 3 (6.42 g). During Experiment 5, a maximum g-force of 8.27 g occurred at impact on the Apple Watch Series 7 Bluetooth.

In Experiment 5, Replicate 1, it was found that the zero-turn mower did not roll down the hill in a manner that would be expected of side upset if an operator had been driving the mower at the time of upset occurrence. To simulate a moving side-upset, the mower was dropped from skid steer forks at a height of 1.5 and 3 meters. However, simulating upsets from a drop instead of driving would not produce identical results. Future research will look at controlling the tractor and mowers used in Experiments 4 and 5 remotely to better simulate a real-world upset.

Based on the findings of this study, the current test procedure did not reliably demonstrate the feasibility of using the tested smart devices to detect agricultural-related incidents. The inconclusive results indicate that further refinement of the testing methodology is necessary before a final recommendation can be made on the effectiveness of these technologies for alerting medical help to lone workers in remote agricultural areas.

Recommendations

Based upon a review of relevant literature and an analysis of the simulated incident data collected during the experiments performed in this study, the following recommendations were developed:

Acknowledgments

This article was made possible by support from Purdue University’s Agricultural Safety and Health Program and USDA/NIFA Special Project 2021-41590-34813.

References

Apple Inc. (2023a). Sensors Toolbox - Multitool. Retrieved from https://apps.apple.com/us/app/sensors-toolbox-multitool/id1383687190

Apple Inc. (2023b). Use crash detection on iPhone or Apple watch to call for help in an accident. Retrieved from https://support.apple.com/en-us/HT213225#:~:text=Set%20up%20your%20device%20for%20an%20emergency&text=To%20share%20your%20location%20with,Calls%20%26%20SOS%20is%20turned%20on.

Apple Inc. (2023c). Use fall detection with apple watch. Retrieved from https://support.apple.com/en-us/HT208944

Ayers, P., Khorsandi, F., Wang, X., & Araujo, G. (2018). ROPS designs to protect operators during agricultural tractor rollovers. J. Terramech., 75, 49-55. https://doi.org/10.1016/j.jterra.2017.05.003

Choi, K. (2023, April). One-tap device sensor logger in your pocket with watch & heart rate logging. Retrieved from https://www.tszheichoi.com/sensorlogger

Deere and Company. (2023). 324G Skid Steer. Retrieved from https://www.deere.com/en/loaders/skid-steers/324g-skid-steer/

Etienne, A. J., Field, W. E., & Haslett, N. J. (2023). A summary of lone agricultural worker injuries and fatalities. J. Agric. Saf. Health, 29(3), 185-201. https://doi.org/10.13031/jash.15523

Garmin. (2023). Setting up incident detection on a garmin device. Retrieved from https://support.garmin.com/en-US/?faq=RfaXahBWkH8Q7pVFLsuUmA

Grupo One Air. (2022). What is g force? Retrieved from https://www.grupooneair.com/what-is-g-force/#:~:text=Colloquially%20known%20as%20a%20force,on%20an%20object%20or%20individual

Kanoski Bresney. (2022). UPDATE: What are the most common types of tractor accidents? Retrieved from https://www.kanoski.com/personal-injury-attorneys/update-what-are-the-most-common-types-of-tractor-accidents#:~:text=The%20National%20Agricultural%20Tractor%20Safety,common%20type%20of%20tractor%20accident

Langer, S., Dietz, D., & Butz, A. (2021). Towards risk indication in mountain biking using smart wearables. Proc. 2021 CHI Conf. on Human Factors in Computing Systems (CHI EA ‘21).Article 467, pp. 1-7. Association for Computing Machinery. https://doi.org/10.1145/3411763.3451746

Larson, D., & Liljedahl, J. B. (1971). Simulation of sidways overturning of wheel tractors on side slopes. Proc. National Farm, Construction, and Industrial Machinery Meeting (pp. 1-12). Society of Automotive Engineers. https://doi.org/10.4271/710709

Lashkari, C. (2019). Types of sensors in wearable fitness trackers. News-Medical.Net. Retrieved from https://www.news-medical.net/health/Types-of-sensors-in-wearable-fitness-trackers.aspx#:~:text=An%20accelerometer%20sensor%20takes%20inertial,actually%20recorded%20by%20this%20sensor.

Mitchell, B. W. (1969). Prediction and control of tractor stability to prevent overturning. West Lafayette, IN: Purdue University. Retrieved from https://www.proquest.com/docview/302481098?pq-origsite=gscholar&fromopenview=true

Moreschi, C., Da Broi, U., Cividino, S., Gubiani, R., Pergher, G., Vello, M., & Rinaldi, F. (2017). The analysis of the cause-effect relation between tractor overturns and traumatic lesions suffered by drivers and passengers: A crucial step in the reconstruction of accident dynamics and the improvement of prevention. Agriculture, 7(12), 1-13. https://doi.org/10.3390/agriculture7120097

Murphy, D. J., Myers, J., McKenzie Jr., E. A., Cavaletto, R., May, J., & Sorensen, J. (2010). Tractors and rollover protection in the United States. J. Agromed., 15(3), 249-263. https://doi.org/10.1080/1059924X.2010.484309

Myers, J. R., & Snyder, K. A. (1996). Rollover protective structure use and the cost of retrofitting tractors in the United States, 1993. J. Agric. Saf. Health, 1(3), 185-197. https://doi.org/10.13031/2013.19463

Nedjai-Merrouche, I., Saadia, N., RamdaneCherif, A., & Makhlouf, A. (2021). Outdoor multimodal system based on smartphone for health monitoring and incident detection. J. Ambient Intell. Human Comput., 12, 10699–10721. https://doi.org/10.1007/s12652-020-02880-5

Nour, M., Cheng, Y.-H., Ni, J.-Q., Sheldon, E., & Field, W. E. (2021). Summary of injuries and fatalities involving livestock manure storage, handling, and transport operations in seven central states: 1976-2019. J. Agric. Saf. Health, 27(2), 105-122. https://doi.org/10.13031/jash.14343

O’Connor, T. (2018). Rollover Protection Structures can prevent injury/death due to tractor rollovers. Omaha, NE: University of Nebraska Medical Center. Retrieved from https://www.unmc.edu/newsroom/2018/08/27/rollover-protection-structures-can-prevent-injury-death-due-to-tractor-rollovers/

Ohkami, Y. (2003). Encyclopedia of physical science and technology (3rd ed.). Academic Press.

Pascuzzi, S. (2015). A multibody approach applied to the study of driver injuries due to a narrow-track wheeled tractor rollover. J. Agric. Eng., 46(3), 105-114.

Smith, D. W. (2011). Safe tractor operation: Rollover prevention. 1-4. Texas A&M System. Retrieved from https://agsafety.tamu.edu/files/2011/06/SAFE-TRACTOR-OPERATION-ROLLOVER1.pdf

Tomer, D. (2022). What I was missing while using the Kalman filter for object tracking. Towards Data Science - Medium. Retrieved from https://towardsdatascience.com/what-i-was-missing-while-using-the-kalman-filter-for-object-tracking-8e4c29f6b795#:~:text=The%20Kalman%20filter%20is%20an,relations%20between%20the%20estimated%20variables.

U.S. Bureau of Labor Statistics. (2020). 146 fatal work injuries involving tractors occurred in 2018. Washington, DC: U.S. Bureau of Labor Statistics. Retrieved from https://www.bls.gov/opub/ted/2020/146-fatal-work-injuries-involving-tractors-occurred-in-2018.htm

Table 1. Details for each experiment and number of replications performed.
Date	Experiment	# Replicates	Total # of Simulated Incidents
3/24/2023	Experiment 1: UTV ejection testing (Garmin watch, Apple Bluetooth watch)	4	4
3/31/2023	Experiment 2, Replicates 1–4: Forklift vertical drop test (Garmin watch, Apple Bluetooth watch)	4	4
7/21/2023	Experiment 2, Replicates 5–8: Forklift vertical drop test 2 (Garmin watch, Apple cellular watch, Apple Bluetooth watch, Movesense Active Tracker)	4	4
4/28/2023	Experiment 3: Skid steer vertical drop test (Garmin watch, Apple cellular watch)	4	4
6/29/2023	Experiment 4: Side tractor upset testing (Apple cellular watch, Garmin watch, Movesense Active tracker)	6	6
7/27/2023	Experiment 5: Zero-turn mower upset testing (Garmin watch, Apple cellular watch, Apple Bluetooth watch, Movesense Active Tracker)	5	5
Total	5 total experiments performed	27	27

Testing the Feasibility of Selected, Commercially Available Wearable Devices in Detecting Agricultural-Related Incidents

Published in Journal of Agricultural Safety and Health 30(4): 181-204 (doi: 10.13031/jash.15985). Copyright 2024 American Society of Agricultural and Biological Engineers.

Methods

Methods for Incident Detection Software Setup

Methods for Experimental Setup

Results

Conclusions

Limitations

Recommendations

Acknowledgments

References