Date of Award


Access Type

Dissertation - Open Access

Degree Name

Doctor of Philosophy in Aviation


Doctoral Studies

Committee Chair

Dr. Antonio Cortés, Ph.D.

First Committee Member

Dr. Bruce Conway, Ph.D.

Second Committee Member

Dr. Dahai Liu, Ph.D.

Third Committee Member

Dr. Carolina Anderson, Ph.D.

Fourth Committee Member

Dr. Wagdi Habashi, Ph.D.


An archival study was conducted to determine the influence of stall warning system performance on aircrew decision-making outcomes during airborne icing encounters. A Conservative Icing Response Bias (CIRB) model was developed to explain the historical variability in aircrew performance in the face of airframe icing. The model combined Bayes’ Theorem with Signal Detection Theory (SDT) concepts to yield testable predictions that were evaluated using a Binary Logistic Regression (BLR) multivariate technique applied to two archives: the NASA Aviation Safety Reporting System (ASRS) incident database, and the National Transportation Safety Board (NTSB) accident databases, both covering the period January 1, 1988 to October 2, 2015.

The CIRB model predicted that aircrew would experience more incorrect response outcomes in the face of missed stall warnings than with stall warning False Alarms. These predicted outcomes were observed at high significance levels in the final sample of 132 NASA/NTSB cases. The CIRB model had high sensitivity and specificity and explained 71.5% (Nagelkerke R2) of the variance of aircrew decision-making outcomes during the icing encounters. The reliability and validity metrics derived from this study suggest indicate that the findings are generalizable to the population of U.S. registered turbine-powered aircraft.

These findings suggest that icing-related stall events could be reduced if the incidence of stall warning misses could be minimized. Observed stall warning misses stemmed from three principal causes: aerodynamic icing effects, which reduced the stall angle-of-attack (AoA) to below the stall warning calibration threshold; tail stalls, which are not monitored by contemporary protection systems; and icing-induced system issues (such as frozen pitot tubes), which compromised stall warning system effectiveness and airframe envelope protections. Each of these sources of missed stall warnings could be addressed by Aerodynamic Performance Monitoring (APM) systems that directly measure the boundary layer airflow adjacent to the affected aerodynamic surfaces, independent of other aircraft stall protection, air data, and AoA systems. In addition to investigating APM systems, measures should also be taken to include the CIRB phenomenon in aircrew training to better prepare crews to cope with airborne icing encounters. The SDT/BLR technique would allow the forecast gains from these improved systems and training processes to be evaluated objectively and quantitatively.

The SDT/BLR model developed for this study has broad application outside the realm of airborne icing. The SDT technique has been extensively validated by prior research, and the BLR is a very robust multivariate technique. Combined, they could be applied to evaluate high order constructs (such as stall awareness for this study), in complex and dynamic environments. The union of SDT and BLR reduces the modeling complexities for each variable into the four binary SDT categories of Hit, Miss, False Alarm, and Correct Rejection, which is the optimum format for the BLR. Despite this reductionist approach to complex situations, the method has demonstrated very high statistical and practical significance, as well as excellent predictive power, when applied to the airborne icing scenario.