A new approach to modeling aviation accidents

Arjun Harsha Rao, Purdue University

Abstract

General Aviation (GA) is a catchall term for all aircraft operations in the US that are not categorized as commercial operations or military flights. GA aircraft account for almost 97% of the US civil aviation fleet. Unfortunately, GA flights have a much higher fatal accident rate than commercial operations. Recent estimates by the Federal Aviation Administration (FAA) showed that the GA fatal accident rate has remained relatively unchanged between 2010 and 2015, with 1566 fatal accidents accounting for 2650 fatalities. Several research efforts have been directed towards betters understanding the causes of GA accidents. Many of these efforts use National Transportation Safety Board (NTSB) accident reports and data. Unfortunately, while these studies easily identify the top types of accidents (e.g., inflight loss of control (LOC)), they usually cannot identify why these accidents are happening. Most NTSB narrative reports for GA accidents are very short (many are only one paragraph long), and do not contain much information on the causes (likely because the causes were not fully identified). NTSB investigators also code each accident using an event-based coding system, which should facilitate identification of patterns and trends in causation, given the high number of GA accidents each year. However, this system is susceptible to investigator interpretation and error, meaning that two investigators may code the same accident differently, or omit applicable codes. To facilitate a potentially better understanding of GA accident causation, this research develops a state-based approach to check for logical gaps or omissions in NTSB accident records, and potentially fills-in the omissions. The state-based approach offers more flexibility as it moves away from the conventional event-based representation of accidents, which classifies events in accidents into several categories such as causes, contributing factors, findings, occurrences, and phase of flight. The method views aviation accidents as a set of hazardous states of a system (pilot and aircraft), and triggers that cause the system to move between hazardous states. I used the NTSB’s accident coding manual (that contains nearly 4000 different codes) to develop a “dictionary” of hazardous states, triggers, and information codes. Then, I created the “grammar”, or a set of rules, that: (1) orders the hazardous states in each accident; and, (2) links the hazardous states using the appropriate triggers. This approach: (1) provides a more correct count of the causes for accidents in the NTSB database; and, (2) checks for gaps or omissions in NTSB accident data, and fills in some of these gaps using logic-based rules. These rules also help identify and count causes for accidents that were not discernable from previous analyses of historical accident data. I apply the model to 6200 helicopter accidents that occurred in the US between 1982 and 2015. First, I identify the states and triggers that are most likely to be associated with fatal and non-fatal accidents. The results suggest that non-fatal accidents, which account for approximately 84% of the accidents, provide valuable opportunities to learn about the causes for accidents. Next, I investigate the causes of inflight loss of control using both a conventional approach and using the state-based approach. The conventional analysis provides little insight into the causal mechanism for LOC. For instance, the top cause of LOC is “aircraft control/directional control not maintained”, which does not provide any insight. In contrast, the state-based analysis showed that pilots’ tendency to clip objects frequently triggered LOC (16.7% of LOC accidents)—this finding was not directly discernable from conventional analyses. Finally, I investigate the causes for improper autorotations using both a conventional approach and the state-based approach. The conventional approach uses modifiers (e.g., “improper”, “misjudged”) associated with “24520: Autorotation” to identify improper autorotations in the pre-2008 system. In the psot-2008 system, the NTSB represents autorotation as a phase of flight, which has no modifier—making it impossible to determine if the autorotation was unsuccessful. In contrast, the state-based analysis identified 632 improper autorotation accidents, compared to 174 with a conventional analysis. Results from the state-based analysis show that not maintaining rotor RPM and improper flare were among the top reasons for improper autorotations. The presence of the “not possible” trigger in 11.6% of improper autorotations, suggests that it was impossible to make an autorotative landing. Improper use of collective is the sixth most frequent trigger for improper autorotation. Correct use of collective pitch control is crucial to maintain rotor RPM during an autorotation (considering that engines are generally not operational during autorotations).

Degree

Ph.D.

Advisors

Marais, Purdue University.

Subject Area

Aerospace engineering|Systems science

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS