Developing a Computational Model of the Pilot’s Best Possible Expectation of Aircraft State Given Vestibular and Visual Cues Can Onur Master’s Thesis Defence Committee Members Dr. Amy Pritchett (Chair) Dr. Eric Johnson Dr. Santosh Mathan This work is funded by the NASA Aviation Safety Program Overview Introduction and Motivation Objectives 1 – Developing the Model 2 – Verify & Validate the Model Conclusion Spatial Disorientation in Aviation Spatial Disorientation (SD): occurs when a pilot fails to properly sense the aircraft’s motion, position or attitude “The chance of a pilot experiencing SD during their career is in the order of 90 to 100 per cent.” Australian Transport Safety Bureau, 2007 Spatial Disorientation (SD) leads Loss of Control (LOC) SKYbrary Flight Safety Foundation, 1992 3 Loss of Control Accidents Fatal Accidents – Worldwide Commercial Jet Fleet – 2003 Through 2012 ( ) Boeing, 2013 Bateman, et al., 2011 4 Aircraft State Perception and Susceptibility to SD Contributor #1: The Vestibular System Vestibular System Semi-Circular Canals Otolith Human vestibular system evolved in a 1-g environment (walking, running, sitting) Limitations: + Threshold Values (No sensation in case of a sub-t maneuver) + Sensor Dynamics (Signals exponentially decay during a sustained stimulus) 5 Vestibular Illusions in Aviation Limitations are causing illusions (especially when visual cues are lacking) Somatogyral Illusions (mostly due to SCC) (e.g., dead-man’s spiral, leans) 6 Somatogravic Illusions (mostly due to otolith) (e.g., false sensation of pitch) Aircraft State Perception and Susceptibility to SD Contributor #2: The Visual System Flight desk displays are the most reliable source of information (If scanned) Contributor #3: Pilot Knowledge of the Aircraft Dynamics • Pilot expertise through training and experience • Ability to generate internal expectations of the aircraft state based on sensory cues 7 Aircraft State Perception and Susceptibility to SD Contributor #2: The Visual System Flight desk displays are the most reliable source of information (If scanned) Problem 1: How does a pilot incorporate these sensory inputs and the expertise into their Contributorexpectation #3: The Knowledge of theorientation? Aircraft Dynamics of spatial • Pilot Expertise through training and experience • Ability to generate internal expectations of the aircraft state based on sensory cues 8 Countermeasures to SD + Training Simulators “Believe your flight instruments” trainings + Alerting Systems Auditory Tactile Visual + Flight Deck Display Designs NextGen Flight Deck Displays Software/Hardware Enhancements 9 Countermeasures to SD + Training Simulators “Believe your flight instruments” trainings + AlertingProblem Systems 2: How to identify the pilot’s information Auditory Tactile Visual requirements? Problem 3: How to help analyze potential flight deck technology interventions? + Flight Deck Display Designs NextGen Flight Deck Displays Software/Hardware Enhancements 10 Overview Introduction and Motivation Objectives 1 – Developing the Model 2 – Verify & Validate the Model Conclusion Objectives 1. Develop a computational model (Model-Based Observer) to predict the pilot’s best possible expectation of the aircraft state given vestibular and visual cues. 1. Parameterize and verify & validate the model using: Preliminary scenarios (brief turns, banking maneuvers etc.) Empirical data from the literature 12 Overview Introduction and Motivation Objectives 1 – Developing the Model 2 – Verify & Validate the Model Conclusion The Model-Based Observer C 1 Aircraft Dynamics 2 Measurements of Aircraft State 3 Pilot’s “Internal Simulation” of the Aircraft + + 2 y Best Possible Pilot Expectations 4 + residuals v - ⌃ ⌃ ⌃ 14 4 Discrepancy between estimated and actual measurements Discrete Visual Scanning Model x = [ h u v w p q r q{q, f, j }]T Discrete Measurements q{q, f, j } f, q u, v, w h u, v, w + Measurement error (v) for visual scan … Sensor noise Error due to display design (thick needle -> elevated error) Pilot perception error + MBO stable for a range of error values 15 timeline Vestibular Model Continuous measurements of the aircraft states and state derivatives The SCC Dynamics (based on Merfeld’s model) ySCC = p * q r The Otolith Dynamics (based on Grant & Best’s model) yOTO = * + Measurement error (v) for vestibular model + Error values given in the previous work Merfeld 1990, Grant & Best 1986 16 MBO Structure: Hybrid Kalman Filter + Continuous-time non-linear system dynamics (aircraft dynamics) P -> the error covariance matrix (a measure of the estimated accuracy of the state estimate) P+ = (I - Kdk Cdk )P- (I - Kdk Cdk )T + Kdk Rdk K Tdk (Discrete RE) · P = AP + PAT + GQGT - PCcT Rc-1Cc P (Continuous RE) Kdk = P-CdTk (Cdk P-CdTk + Rdk ) -1 (Discrete Kalman Gain ) Kc = P-CcT (Rc ) -1 (Continuous Kalman Gain) Kd & Kc are the optimal gains (Kalman Gains) to generate the best possible estimate 17 Overview Introduction and Motivation Objectives 1 – Developing the Model 2 – Verify & Validate the Model Conclusion Verification of the Model Components i. ii. iii. iv. 19 The Hybrid Kalman Filter The Semi-Circular Canal Model The Otolith Model The Dryden Model The MBO components Implemented in aircraft simulation to emulate turbulence i) Kalman Filter Verification Gaussian Error is used in the Kalman Filter Estimation error is expected to be Gaussian Kalman Entries filter estimate= Diagonal SCC afferences Predicted estimation actual error variances of each state q - value (rad/s) 0.1 0.05 0 -0.05 -0.1 0 5 10 15 20 time (s) 25 30 35 q - error value 0.1 +2sigma -2sigma Estimate error 0.05 0 -0.05 -0.1 20 40 0 5 10 15 20 time (s) 25 30 35 40 i) Kalman Filter Verification Gaussian Error is used in the Kalman Filter Estimation error is expected to be Gaussian q (roll rate) - estimation error distribution 450 Kalman Entries filter estimate= Diagonal SCC afferences Predicted estimation actual error variances of each state 0.1 350 q - value (rad/s) Mean = -0.00046 Std. Dev. = 0.00788 Median = -0.00015 400 300 250 0.05 0 -0.05 -0.1 200 0 5 10 15 150 20 time (s) 25 30 35 0.1 50 0 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 q - error value 100 +2sigma -2sigma Estimate error 0.05 0 -0.05 -0.1 21 40 0 5 10 15 20 time (s) 25 30 35 40 q - value (rad/s) 0.1 Kalman filter estimate actual 0.05 i) Kalman Filter Verification – Measurements Impact 0 -0.05 -0.1 0 5 10 15 20 time (s) 25 qq--value error value (rad/s) 0.05 0.05 40 actual Sub-Threshold (no visual – no SCC) 0 0 -0.05 -0.1 0 5 10 15 20 time (s) time (s) 25 30 35 40 +2sigma -2sigma Estimate error 0.1 q - error value 35 +2sigma -2sigma Kalman filter estimate Estimate error 0.1 0.1 0.05 Above-Threshold (no visual) 0 -0.05 -0.1 22 30 0 5 10 15 20 time (s) 25 30 35 40 ii) SCC Model Verification Angular Velocity (rad/s) 0.25 Stimulus (step) Canal Afferent Response 0.2 0.15 0.1 0.05 0 -0.05 0 5 10 15 20 25 Time (s) 30 35 40 45 0.12 Stimulus (Actual Aircraft State) Canal Afferent Response 0.1 Ang. Velocity (p) [rad/s] 0.08 A previously developed model’s responses (Borah et. al, 1988) 0.06 0.04 0.02 0 -0.02 -0.04 10 23 15 20 25 30 Time [s] 35 40 45 50 ii) SCC Model Verification (sub-threshold behavior) + The SCC does not provide accurate information in case of a sub-threshold maneuver Does it provide no measurement? (is it completely inactive) Does it provide measurements of zero? 24 p - value (rad/s) Kalman filter estimate actual 0.05 0 ii) SCC Model Verification (sub-threshold behavior) -0.05 -0.1 0 (1) The SCC provides of 20 zero when25 maneuver30 is sub-threshold 5 10 measurements 15 35 40 time (s) +2sigma -2sigma Estimate error p - value (rad/s) p - error value 0.1 0.05 0.1 Pilot Expectation Aircraft State 0 0.05 -0.05 0 -0.1 -0.05 0 -0.1 0 5 10 15 20 time (s) 25 30 35 10 15 20 25 30 35 (2) The5SCC provides no measurement when maneuver is sub-threshold 40 40 time (s) +2sigma -2sigma Estimate error p - error value 0.1 0.05 0 -0.05 -0.1 0 5 10 15 20 time (s) Sub-Threshold Above-T Sub-Threshold 25 25 30 35 40 iii) The Otolith Model Verification -4 2 x 10 1 otolith afferent firing rate 0 sf x -1 Forward Acceleration Experiment -2 -3 -4 -5 sf_x = –θ.g – ů -6 0 10 15 Kalman filter estimate actual 30 35 40 9 8 0 7 0 5 10 15 20 time (s) 25 30 100 theta - error value 25 5 -5 35 40 +2sigma -2sigma Estimate error 50 6 5 4 0 3 -50 -100 26 20 time (s) linear acceleration theta - value (deg) 10 5 2 0 0 5 10 15 20 time (s) 25 30 35 40 5 10 15 20 time (s) 25 30 35 40 iii) The Otolith Model Verification -4 6 x 10 otolith afferent firing rate 4 sf x 2 0 -2 Pitch-up Experiment -4 sf_x = –θ.g – ů -6 0 5 30 35 40 4 3 2 0 0 5 10 15 20 time (s) 25 30 35 40 +2sigma -2sigma Estimate error 50 linear acceleration theta - value (deg) 25 5 100 theta - error value 20 time (s) 10 -5 27 15 Kalman filter estimate actual 15 (While we didn’t command a deceleration the pitch up did cause it) 10 1 0 -1 -2 0 -3 -50 -4 -100 0 5 10 15 20 time (s) 25 30 35 40 -5 0 5 10 15 20 time (s) 25 30 35 40 iv) The Dryden Model Verification WMC Dryden Implementation V - linear component gusts magnitude spectrum Vz 0 Magnitude Vy 20 Linear gust verification 800 Vx WMC Dryden Implementation 600 400 200 -20 0 0 1 2 3 4 5 6 7 8 9 10 Magnitude Vz 0 -20 2 3 4 5 6 40 60 80 Frequency (bins) Matlab Dryden Built-in 100 120 0 20 40 60 80 Frequency (bins) 100 120 800 Vy 20 1 20 Vx Matlab Dryden Built-in 0 0 7 8 9 600 400 200 0 10 WMC Dryden Implementation WMC Dryden Implementation wx 0.05 wy 0 wz -0.05 Angular gust verification W - angular component gusts magnitude spectrum 1 Magnitude 0.1 0.5 -0.1 0 0 1 2 3 4 5 6 7 8 9 10 0 20 40 60 80 Frequency (bins) Matlab Dryden Built-in 100 120 0 20 40 60 80 Frequency (bins) 100 120 Matlab Dryden Built-in wx 0.05 wy 0 wz 1 Magnitude 0.1 -0.05 0.5 -0.1 0 0 28 1 2 3 4 5 6 7 8 9 10 Validation of the Integrated MBO Ability of the model to predict known problems with pilot SD. Specifically, to reproduce the illusions that occur due to vestibular limitations when visual cues are lacking. 29 Impact of visual scanning. Do the visual corrections help overcome the illusion? Somatogyral Illusion Above-Threshold Banking Sub-Threshold Banking 0.05 0.05 00 -0.05 -0.05 -0.1 -0.1 00 55 10 10 15 15 20 20 time(s) (s) time 25 25 35 35 0.05 0.05 00 -0.05 -0.05 -0.1 -0.1 0 0 30 5 5 10 10 15 15 20 20 time (s) time (s) 25 25 30 30 35 35 40 40 Kalman filter estimate actual 0.05 0 -0.05 -0.1 40 40 +2sigma -2sigma +2sigma Estimate error -2sigma Estimate error 0.1 0.1 pp- -error errorvalue value 30 30 p - value (rad/s) 0.1 Kalmanfilter filterestimate estimate Kalman actual actual 0 5 10 15 20 time (s) 25 30 35 40 +2sigma -2sigma Estimate error 0.1 p - error value pp--value value (rad/s) (rad/s) 0.1 0.1 0.05 0 -0.05 -0.1 0 5 10 15 20 time (s) 25 30 35 40 Visual Corrections on Somatogyral Illusion Above-Threshold Banking Pilot Expectation Aircraft State 0.05 0 -0.05 -0.1 5 10 15 20 time (s) 25 30 +2sigma -2sigma Estimate error 0.05 0 -0.05 31 0 -0.05 40 0 5 10 15 20 time (s) 25 30 35 40 0 5 10 15 20 time (s) 25 20 time (s) 25 30 35 40 +2sigma -2sigma Estimate error 0.1 p - error value p - error value 35 Pilot Expectation Aircraft State 0.05 -0.1 0 0.1 -0.1 0.1 p - value (rad/s) p - value (rad/s) 0.1 Sub-Threshold Banking 0.05 0 -0.05 -0.1 0 5 10 15 30 35 40 Somatogravic Illusion Forward Acceleration Pilot Expectation Aircraft State 9 0 -5 8 0 5 15 50 25 30 35 40 7 6 5 4 3 0 x-component (ft/sec 2) 2 -50 -100 1 0 5 10 15 20 time (s) 25 30 35 0 10 15 20 time (s) 25 30 35 40 40 Deceleration 0 3.5 3 -5 -10 2.5 2 0 5 100 10 15 20 time (s) 25 30 35 40 +2sigma -2sigma Estimate error 50 1.5 1 0.5 0 0 -0.5 -1 -50 -1.5 -100 32 5 Pilot Expectation Aircraft State 5 theta - value (deg) 20 time (s) linear acceleration theta - error value 10 +2sigma -2sigma Estimate error 100 theta - error value 10 5 linear acceleration theta - value (deg) 10 0 5 10 15 20 time (s) 25 30 35 40 0 5 10 15 20 time (s) 25 30 35 40 Visual Corrections on Somatogravic Illusion Forward Acceleration Pilot Expectation Aircraft State 5 9 0 8 -5 7 -10 0 5 10 15 20 time (s) 25 30 40 +2sigma -2sigma Estimate error 20 theta - error value 35 10 linear acceleration theta - value (deg) 10 6 5 4 0 3 -10 2 -20 33 10 0 5 10 15 20 time (s) 25 30 35 40 1 x-component (ft/sec 2) 0 5 10 15 20 time (s) 25 30 35 40 Overview Introduction and Motivation Objectives 1 – Developing the Model 2 – Verify & Validate the Model Conclusion Examples & Potential Design Interventions Concerns Model Representation Potential Design Intervention Pilot distraction No scan for some or all of the instruments Alerting for rare (and rarely-sampled) flight conditions. Inaccurate pilot perception of state from instruments Inaccurate or noisy measurement More accurate/higher resolution presentation of key aircraft states. 35 Summary & Contributions The MBO enables several analyses: Investigate the mechanism of spatial disorientation Predict the best possible pilot’s expectations of the aircraft state with a given visual scan pattern Identify the pilot’s information requirements (regarding the appropriate energy-state and attitude awareness) Analyze potential flight deck technology interventions and/or provide design insights for the NextGen flight deck display designs. 36 Thank You! 37