1-2 of 2 results
-
Examining Unstable Approach Predictors Using Flight Data Monitoring Information
PI David Carroll
CO-I David Esser
The approach and landing phase of flight is statistically the most dangerous part of flying. While it only accounts for 4% of flight time, it represents 49% of commercial jet mishaps. One key to mitigating the risks involved in this flight segment is the stabilized approach. A stabilized approach requires meeting rigorous standards for many flight parameters as the aircraft nears landing. Exceeding any of these parameters results in an unstable approach (UA). The energy management (EM) accomplished by the flight crew, represented by the EM variables in the study, influences the execution of a stabilized approach.
While EM is a critical element of executing a stabilized approach, there appears to be a lack of studies that identify specific EM variables that contribute to UA probability. Additionally, several possible moderating variables (MVs) may affect the probability of a UA. Fortunately, modern jet transport aircraft have Flight Data Monitoring (FDM) systems that capture a wealth of information that enable the analysis of these EM variables. This study used FDM data to answer the questions about what influence a set of EM variables has on the probability of a UA event. The analysis also determined what impact a set of possible MVs, not directly related to EM, has on these EM variables influence.
The analysis used logistic regression (LR) to investigate FDM information. The LR provided estimations of odds ratios for each of the variables and the interaction factors for the MVs. These statistics defined a model to evaluate the influences of the EM and MVs, providing answers to the research questions posed. The results determined the model was a good fit to the data but had poor discrimination. The model supported three of the original seven EM hypotheses and none of the 28 MV hypotheses.
The study identified three specific EM variables that significantly influenced the probability of a UA event. Of the MVs, only one significant influence was revealed but was opposite that hypothesized. Identifying the EM variables, and examining their impacts, shows their importance in preventing UAs. Further, the results help prevent future UAs by informing the design of training programs. Additionally, the current effort fills gaps in the current body of knowledge, as there appears to be a lack of studies in the areas investigated.
A gap in the body of knowledge filled by investigating an area of limited research and the results provide practical application in the analysis of EM-related events. Aviation safety practitioners now have additional information to identify trend issues that may lead to the increased probability of a UA event. Finally, this study was one of very few granted access to actual operational FDM information by an air carrier. The data were crucial in evaluating the proposed model against real-world flight operations, comparing theory to reality. Without access to such closely held information, the research for this dissertation would not have been possible.
Categories: Graduate
-
Predicting Pilot Misperception of Runway Excursion Risk Through Machine Learning Algorithms of Recorded Flight Data
PI Edwin Odisho
CO-I Dothang Truong
The research used predictive models to determine pilot misperception of runway excursion risk associated with unstable approaches. The Federal Aviation Administration defined runway excursion as a veer-off or overrun of the runway surface. The Federal Aviation Administration also defined a stable approach as an aircraft meeting the following criteria: (a) on target approach airspeed, (b) correct attitude, (c) landing configuration, (d) nominal descent angle/rate, and (e) on a straight flight path to the runway touchdown zone. Continuing an unstable approach to landing was defined as Unstable Approach Risk Misperception in this research. A review of the literature revealed that an unstable approach followed by the failure to execute a rejected landing was a common contributing factor in runway excursions.
Flight Data Recorder data were archived and made available by the National Aeronautics and Space Administration for public use. These data were collected over a four-year period from the flight data recorders of a fleet of 35 regional jets operating in the National Airspace System. The archived data were processed and explored for evidence of unstable approaches and to determine whether or not a rejected landing was executed. Once identified, those data revealing evidence of unstable approaches were processed for the purposes of building predictive models.
SAS™ Enterprise MinerR was used to explore the data, as well as to build and assess predictive models. The advanced machine learning algorithms utilized included: (a) support vector machine, (b) random forest, (c) gradient boosting, (d) decision tree, (e) logistic regression, and (f) neural network. The models were evaluated and compared to determine the best prediction model. Based on the model comparison, the decision tree model was determined to have the highest predictive value.
The Flight Data Recorder data were then analyzed to determine predictive accuracy of the target variable and to determine important predictors of the target variable, Unstable Approach Risk Misperception. Results of the study indicated that the predictive accuracy of the best performing model, decision tree, was 99%. Findings indicated that six variables stood out in the prediction of Unstable Approach Risk Misperception: (1) glideslope deviation, (2) selected approach speed deviation (3) localizer deviation, (4) flaps not extended, (5) drift angle, and (6) approach speed deviation. These variables were listed in order of importance based on results of the decision tree predictive model analysis.
The results of the study are of interest to aviation researchers as well as airline pilot training managers. It is suggested that the ability to predict the probability of pilot misperception of runway excursion risk could influence the development of new pilot simulator training scenarios and strategies. The research aids avionics providers in the development of predictive runway excursion alerting display technologies.
Categories: Graduate
1-2 of 2 results