1998 - Draper Laboratory

Notice that each flight track described in Table 3 is beingmodeled as one or more vectors of the five functionalcomponents. Once this association is defined, the Markov modelsupplies the probabilities of the vector components. Theresulting probability of the vector is the product of thecomponent probabilities, and in the case where a flight track isassociated with several vectors, such as ‘fake,’ the associatedprobability is the sum of the vector probabilities (a numericalevaluation is given in Table 6).case in which an alert is issued, but is too late to prevent acollision.Table 4. Outcome categories.Interaction-Response ModelThe Interaction-Response Model used in this study wasoriginally developed at MIT (Refs. [5], [8]). The performance ofthe prototype alerting system is evaluated using the piloted flightsimulation tracks (Ref. [6]) described above. In the evaluations,the threatened aircraft, the evader, always follows a normalapproach path while the intruder follows one of the blunder ornormal approach paths from the simulation tests. The alertinglogic is implemented only on the evader. If an alert is issued, theevader performs a specified climbing-turn avoidance maneuver.The outcome of each approach is recorded according to thepossible outcomes listed in Table 4. A collision is defined tooccur if separation at any point in the approach is less than500 ft.From Table 4, if an alert is not issued at any time during a run,it is classified as either a Correct Rejection (if a collision did notoccur) or as a Missed Detection (if a collision did occur). If analert is issued, the outcome is placed in one of four categories.An Unnecessary Alert is a case where the intruder is not on acollision course, an alert is issued anyway, and a collision is stillavoided. If a collision occurs because of the alert, it is classifiedas an Induced Collision. A Correct Detection occurs when acollision is averted because of an alert. Finally, a Late Alert is aThe outcome categories of Table 4 can be combined to yieldthree safety statistics defined as follows:ProbabilityofReliableOperation=Probability of Collision =Probability of False Alarm =# Correct Rej. + # Correct Det.Total # of Runs(3-1)# Mis. Det. + # Ind. Col. + # Late AlertsTotal # of Runs# Unnecessary AlertsTotal # of Runs(3-2)(3-3)In this evaluation, eight pairs of flight tracks were evaluated atthe 1700-ft runway spacing. A total of 183 runs were made foreach pair of tracks, and the ownship was flying the normalFlight Tracknorm: Normal approach to landing [N1,~S3,A1,G1,P1]Markov Model Subvectorsfake:oadj:Aircraft fakes blunder toward othership'srunwayAircraft drifts away from own and other'srunway and then overadjusts[N4,~S3,A1,G1,P1];[N1,~S3,A3,G1,P1];[N1,~S3,A3,G1,P2];[N1,~S3,A3,G2,P1][N4,~S3,A1,G1,P1];[N1,~S3,A3,G1,P1];[N1,~S3,A3,G1,P2];[N1,~S3,A3,G2,P2]sb5: Constant 5-deg bank angle blunder [N2,~S3,A1,G1,P1]; [N1,~S3,A1,G1,P2];[N2,~S3,A1,G2,P1]sh5: Slow 5-deg heading change blunder [N2,~S3,A1,G1,P1];[N1,~S3,A1,G1,P2];[N2,~S3,A1,G1,P2]slo: Slow 10-deg heading change blunder [N2,~S3,A1,G1,P1];[N1,~S3,A1,G1,P2];[N2,~S3,A1,G2,P2]bl15: 15-deg heading blunder [N1,~S3,A1,G3,P1];[N1,~S3,A1,G1,P3];[N3,~S3,A1,G1,P1]bl30: 30-deg heading blunder [N1,~S3,A1,G3,P1];[N1,~S3,A1,G1,P3];[N3,~S3,A1,G3,P3]Figure 8. Impact model mapping.An Integrated Safety Analysis Methodology for Emerging Air Transport Technologies8