In today’s world, technological innovations are occurring at an increasingly rapid pace, and alongside those emerging technologies often come emerging risks. For companies to understand and mitigate these risks, they must first be quantified, but unfortunately, company-specific claim data is typically not available. Thus, the search for relevant industry data and research becomes a key part of the quantification process. As part of SIGMA’s continuous effort to help our clients and partners thrive in a data-driven world, we wanted to use today’s blog as an opportunity to highlight one such source of potentially valuable research.
In 2017, SIGMA was contacted by Chris Duling, a research engineer at the University of Alabama in Huntsville, to discuss his team’s research related to UAS (Unmanned Aircraft Systems) risks. From this initial contact grew an ongoing dialogue on how their extensive research and data compilation might assist companies that are either looking to insure their own UAS-related risks or seeking ways to underwrite third-party UAS risk.
The research conducted by Chris and his team focuses on the risk of a UAS causing either property damage or bodily injury in the event of unplanned ground collision. The use of UAS has already spread across numerous industries and is clearly on a path to continue growing, so we believe their research could be valuable to many in the insurance and risk management arena.
For the remainder of this blog, Chris and his team will be answering various questions regarding emerging UAS risk. We want to be clear that the research being discussed did not involve any participation from SIGMA, and the following interview is strictly for the purposes of allowing Chris, Nishanth Goli, and the rest of the UAH team to discuss their findings and expand upon the ways their research may be useful.
To get us started, what type of companies or industries in the risk management and insurance arena do you think would most benefit from your research?
Currently, the UAS market has three groups that can benefit from our research:
i) UAS service providers (for example: precision land management, surveying, etc.) who support various UAS operations for others,
ii) Direct end users (for example: media companies, public safety, etc.) who use UAS for their own operations, and
iii) Commercial UAS manufacturers who make UAS for others (for example: DJI, Parrot, Skydio, etc.) or themselves (Amazon Prime Air, etc.).
Each of these companies/industries must consider the risk of flight safety to people and infrastructure on the ground. The FAA is concerned with these same hazards and only allows commercial flight over people following approval of waiver requests to Part 107 rules that prohibit flight operations over people. Hobbyists are supposed to comply with community-based standards, like the rules of the Academy of Model Aeronautics, which prohibits flight over people.
UAH developed a modeling approach that can be applied to unmanned operations where flight over humans and/or infrastructure is desired. It can quantify the risk using high-fidelity UAS dynamics modeling, human population modeling, and infrastructure modeling. Not only is the maximum probability of sUAS-human impacts evaluated for a given mission, one can estimate how risk is modified when the mission is modified. This research is valuable to all in the UAS market who desire to conduct operations over people or near critical infrastructure.
Can you summarize the history and current status of your research, including other institutions that are involved?
In May 2015, the FAA Alliance for Systems Safety of UAS through Research Excellence (ASSURE) was designated as the Center of Excellence for UAS research. The University of Alabama in Huntsville (UAH) is a core member of FAA ASSURE and leads research work on Ground Collision Severity in collaboration with other universities and industry partners. This has led to three research projects: Task A4: UAS Ground Collision Severity Evaluation 2015-2017, Task A11: Part 107 Waiver Request Case Study, and Task A14: UAS Ground Collision Severity Evaluation 2017-2019. Six tasks were performed as part of phase II of the Ground Collision Severity research from 2017-2019. These are sUAS failure flight testing, sUAS dynamic modeling and simulation, simplified impact testing, sUAS-Anthropomorphic Test Dummy (ATD) collision tests, Finite Element Analysis (FEA) modeling of sUAS-human collision, and sUAS Post Mortem Human Surrogate (PMHS) collision testing.
The sUAS failure flight testing involved flying sUAS between 350-400 feet altitude and forcing different types of failures. The vehicle would fall for approximately 4 seconds and then was recovered using a parachute. The vehicle state data (position, velocity, altitude, and angular rates) were recorded during the fall. The primary use of this data was to determine relevant ATD and PMHS impact test points based on the unique aerodynamics and dynamics of each tested vehicle. However, the research was also used to develop 6DOF flight dynamic models for each vehicle. These dynamic models were validated with the failure flight tests and can estimate the vehicle trajectory and impact characteristics post mid-air failure under a range of ambient conditions and velocity/altitude combinations, as well as types of failure (1-motor, 2-motor, 4-motor, etc.).
UAH conducted a Monte Carlo Simulation of an individual aircraft’s post-failure dynamics by considering various types of failures, different vehicle states of the flight profile, and environmental variables (wind, etc.). The simulation resulted in North and East horizontal displacement of the vehicle from point of failure to impact. If we can simulate the locations of non-participating public within the region, we can predict the probability of an impact and, based upon our impact studies, we can assess the severity of the impact to develop a risk management approach to various operations using any type of vehicle.
There are two different components to the liability associated with UAS and ground collision: the probability of property damage and the probability of bodily injury. How does your research address these components from a quantification standpoint?
Our research was developed from understanding the severity of impact from a wide variety of drones and was coupled with simulations developed from flight tests. To complete the assessment of risk to the non-participating public, UAH has focused its recent internally-funded research on an application where humans follow a randomized but predictable movement routine within specified terrain based upon an activity on the ground. This approach was applied to evaluating the risk of serious injury to non-participating people whom are under or near a UAS flight path. A high-fidelity model could therefore be developed for any ground activity being monitored by a UAS to determine how to minimize the risk to the non-participating public. The extension of this research to developing modeling for a wide range of fixed infrastructure is quite feasible. Even vehicular movement, which can seem random in an urban population, follows a predictable movement routine and can be transitioned to models. In dense human populations, certain scenarios will allow for population modeling to identify risk areas and risk-free areas. Estimation of the probability of impact can be coupled with the probability of bodily injury using the testing methodology developed by UAH and delivered to the FAA. The approach is currently the basis for ASTM Standard F3889-20 Standard Test Method for Assessing the Safety of Small Unmanned Aircraft Impacts.
Our work under FAA Tasks A4, A11, and A14 more directly addresses the severity of bodily injury using ATD impact tests and estimation of the probability and severity of head and neck injuries using automotive-based injury risk curves. Evaluation of the injury potential of aircraft must be done on a case-by-case basis, as there are dramatically different shapes and materials used to construct different unmanned systems. The geometry and weight of the aircraft determine its aerodynamics and maximum pre-impact kinetic energy, and the materials and structure determine the severity of impact based on the amount of kinetic energy it brings to an impact. Softer, deformable aircraft like foam-based fixed wing aircraft with a pusher propeller have proven, in our testing, to have a much lower risk for human injury than heavy plastic and composite multi-rotor aircraft.
How might your research be used from a risk mitigation standpoint? Why is the ability to quantify UAS-related risk such an important part of this?
The impact testing we’ve done under the FAA research programs is useful for quantifying the injury potential of specific aircraft and determining if mitigations must be applied, e.g. parachute recovery systems to slow descent or lowering the operating altitude to limit the maximum achievable impact velocity in a fall. The current approach being used in the NPRM provided by the FAA assumes a collision with the non-participating public will always occur!
Modeling that incorporates simulation of large numbers of failures can be used to determine the hazard area of the aircraft at any point during a flight, i.e. the area that it can actually impact if it fails at any point in flight along its prescribed trajectories. When the probability of impact at any point is used in conjunction with models of human or vehicular movement in an operating area, then the joint probability of an unmanned system and person being collocated at any point can be calculated. The modeling effort also enables researchers to focus on critical phases of flight to modify the vehicle flight path and quantifiably reduce the probability of impact to a person or critical infrastructure.
Probability of impact, along with the probability and severity of an injury for a specific aircraft flying a specific mission, enables a more complete understanding of hazards and the likelihood of occurrence of these hazards. This understanding can then be included in an overall operational risk assessment to minimize risk to the non-participating public.
Can you briefly summarize some of the issues you dealt with when compiling the data used in your models? For risk management professionals seeking to compile data for their own emerging risks, do you have any suggestions for ways in which these issues can be alleviated?
First and foremost, development of aerodynamic models and relevant impact tests requires failure flight testing of the aircraft. In the absence of that flight testing, it is challenging to accurately estimate the most probable worst-case type impact conditions that a specific aircraft may have. Different aircraft fall at different velocities and behave differently after failures. From our previous research, a DJI Phantom 3 does not fall much faster than 60 ft/s, whereas a DJI Inspire 2 will fall somewhat faster based on its weight and more streamlined body. The DJI Mavic Pro appeared to be able to slow its descent if only one motor failed, which was not at all seen with the other aircraft that we flew for the FAA research.
A quantifiable risk model requires a high-fidelity approach that is only applicable for that scenario or similar scenarios. The modeling required to quantify the risk of a UAS used in agricultural applications will be different than the risk during an urban news coverage using the same UAS. The difference results from the human populations near these operations and the difference in flight profiles and terrain or obstacles located in each environment. Therefore, risk management professionals must consider the work required to compile UAS-specific data, the environmental data where the operations will occur, and the temporal and physical movement of people and critical infrastructure on the ground for each scenario.
Here, we develop two models: a UAS model and the environment/population model. The method required to develop the former model has been developed at UAH and can be extended to other UAS. However, developing an environment/human population model for every application can be challenging. One must involve the various stakeholders to identify the parameters required to model the overall pattern of human movement, as well as develop flight trajectories that may be involved to accomplish the mission given the environment/obstacles that may exist. When developing our human population model for a golf course precision land management application, we evaluated golf play and its statistics for the specific site including human behavior on the course, the driving range, and even the clubhouse and parking lot. UAH worked with the golf course and UAS operator to understand the layout and statistics of the particular golf course, as well as the flight profiles and data that was required to be collected to conduct the mission. All of this information combined into a movement model to predict a player’s position from the time they arrived at the clubhouse, went to the range and putting green to practice, played a round, and returned to the parking lot to return home.
Can you elaborate on the ways you developed the assumptions used in your research? How might others go about developing or selecting assumptions during the quantification process?
Assumptions were mostly made while developing the human population model. The assumptions were dependent on the amount of information that was available. Humans are free-thinking and do not follow a robot-like logic when they move from point A to point B. We assumed that a person will not take a longer distance route nor a longer time route nor a route which they cannot walk/drive. Therefore, we assume that a person will take the most direct feasible path while minimizing distance or time where applicable. One can also predict the speed at which a person can walk/drive. These assumptions, when put together, can predict the person’s location on this path. We also considered how long people move around tees and greens versus fairways based upon the use of cart paths, since time in their carts was considered partially protected time. Development of assumptions and movement patterns takes some discussion and monitoring of an operation to obtain reasonable assessments of the human movement in and around the facility. Using data from security cameras, looking at scheduling information, and direct observation are valuable tools in developing assumptions for human movement.
Can you share a practical example of how you have tested your research?
This entire research started with Greensight™, which partnered with UAH to support FAA Task A14. Greensight™ is a precision land management company that provides its services to golf courses for vegetation inspection. A flight testing and modeling effort on their vehicle was done as part of A14. Later, the company shared with UAH its efforts to obtain a flight over people waiver by generating a low-fidelity model to show a low probability of impact. UAH identified where a higher-fidelity modeling approach could better quantify the risk, so we began gathering operational data from Greensight™ to perform the modeling.
We developed a UAS Impact Probability Density map showing the probability of a UAS impacting at any given location on the golf course based on the flight profile of the UAS. We also developed a Player Population Probability Density map that shows the probability of player presence at any given location on the golf course based on a predictable pattern of player movements on the golf course. When we combine these two maps, we arrive at the UAS-Player Collision Probability Density Maps. Here, at any given location on the golf course, we have an estimate of the probability of a UAS-Human Collision. Additionally, we were able to identify the changes in flight profile that could considerably lower the probability of a UAS-Human Collision.
Can you provide links you would like to share regarding your research?
UAH has been a leading researcher of UAS Ground Collision Severity Studies as part of the FAA’s UAS Center of Excellence, ASSURE, since its inception in 2015. UAH has led three studies under the UAS CoE.
Reports related to these research programs are available from ASSURE.
Can you provide contact information for those seeking further details or discussion?
Chris Duling – firstname.lastname@example.org (reach out to Chris for any questions regarding collaboration)
Nishanth Goli – email@example.com (reach out to Nishanth for any questions regarding the research and the technical approach)
We welcome your feedback by posting a comment or contacting us at support@SIGMAactuary.com.
© SIGMA Actuarial Consulting Group, Inc.