The Use of Pilot Models to Support Flight Simulation: The Sky is the Envelope

Transcription

1 The Use of Pilot Models to Support Flight Simulation: The Sky is the Envelope by Ruud Hosman AMS Consult, the Netherlands Abstract: The first attempts to develop pilot models started after the Second World War. This led in the sixties to a landmark publication by McRuer on Human Pilot Dynamics. While the first pilot models were based on visual information only as pilot input, the introduction of the 6DOF motion systems for flight simulation stimulated experimental research which broadened the understanding and further development of pilot models. Since, there is a direct relation between the application of pilot models and simulation. Pilot models were extended with descriptions of the visual, vestibular and the neuromuscular systems and are ready to be used for a wide range of applications. The paper discusses three applications of pilot models to support flight simulation: pre-experimental analysis, optimization of the cueing algorithm parameters, and analysis of pilot s control performance in the balked landing. The results of these analyses demonstrate that pilot models became mature and are ready to be used for analysis to support flight simulation. 1. Introduction The goal of developing pilot models is threefold: The first is to understand how pilots generate their control output and express that understanding in a mathematical description. The second is to be able to quantify pilot s control behavior by matching the pilot model parameters to measured pilot control behavior. This allows quantifying the changes in pilot s behavior due to different control tasks. The third is that pilot models can be used to perform analysis for many applications. The application of pilot models for analysis of piloted simulations is the topic of this paper Figure 1. Pilot-aircraft control loop with visual and vestibular feedback and external disturbances; a forcing function i(t) and an external disturbance w(t). 1

2 The first attempts to develop pilot models started after the Second World War. This led in the sixties to a landmark publication by McRuer on Human Pilot Dynamics (McRuer et al., 1965). Initially visual information presented to the pilot on an instrument or display was the only input to the pilot model. After the introduction of the 6 DOF motion systems for simulation, the contribution of motion feedback on pilot s control behavior could be investigated and incorporation of a vestibular model into the pilot models became possible and pilot models were extended and further developed, Fig. 1. Since then there has been an interaction between pilot model development and simulation. Today, we have several pilot models available which take visual and vestibular input to the pilot into account and may have an advanced neuromuscular system model to describe the force/displacement interaction between the pilot and the controls. To perform an analysis based on a pilot model to support flight simulation one first has to select the appropriate pilot model to perform the task. The choice of the pilot model depends on the control task configuration to be analyzed. Pilot models are mostly developed to perform the inner loop attitude control task and have to be extended for the more complicated nested loop configuration wherein the pilot performs his manual control task in the daily flight operation. Three models are mentioned here which is definitely not meant as a complete overview of pilot models. The optimal Control model (Kleinman and, Baron, 1973), the structural pilot model (Hess, 1980) and the descriptive pilot model (Hosman, 1996) are pilot models which has been used extensively for pilot vehicle analysis. Figure 2. The descriptive pilot model (Hosman, 1996). In this paper three examples of pilot model analysis will be discussed. The pilot model used for these analyses is the descriptive model (Hosman, 1996) which has visual and vestibular feedback input, Fig. 2. This model is based on extensive experimental research. The only free parameters to be determined are the gains K i weighting the contribution of the sensory outputs R i of the visual and vestibular sensors. These gains are optimized for each control configuration, i.e control task, aircraft configuration, external disturbance, etc.. When pilots adjust their control behavior to a certain control task, they will optimize their behavior to achieve: Good tracking performance Effective control effort 2

3 Adequate bandwidth and stability of the control loop To achieve this goal with the pilot model, the free model parameters are adjusted in an optimization process with the following cost function: J = ( e + Q. u + R. u& ) (1) By selecting the weighting factors Q and R the pilot model control behavior can be adjusted to the proper bandwidth. For which kind of analysis pilot models can be used? In the next sections three examples will be discussed which have a direct relation to flight simulation, i.e. pre experimental analysis, motion cueing algorithm adjustment, and a flight performance analysis. 2. The use of pilot models for pre-experimental analysis. Experimental research with simulators is mostly expensive. The yearly number of hours of research simulators is mostly low compared to training simulators, and experiments have to be run with enough subject pilots to obtain the required statistical power which dictates the number of simulator hours. A thorough preparation and design of the experiment (selection of the experimental configurations, nullhypothesis, number of subjects required, statistical power, etc.) is required to make the experiment effective. A pre-experimental analysis can support and optimize the design of an experiment. As an example of an analysis, the following experiment and the results are discussed below. In the 90th a series of experiments to support the development of helicopter flight simulation motion platform requirements was performed at NASA Ames (Schroeder 1999). Experiments with yaw, roll/lateral Figure 3. The pilot location in plan view. and vertical control tasks were performed. All the experimental details were accurately described which formed a good basis for the analysis. Based on a suggestion of Jeff Schroeder a pilot model analysis was performed on the yaw control tasks and the results presented during the Motion Workshop of the AIAA MST Conference in 2003 (Hosman, 2003). Task set up The main task was a yaw control task with different specifications, i.e. a 15 º yaw capture task, a 180º hover turn task and yaw rotational regulation task. The aircraft dynamics used represented an unaugmented AH-64 Apache helicopter. The model had been identified based on flight test data. The 3

4 aircraft plan view, with the pilot s position relative to the yaw axis at the center of mass (c.m.), is shown in Fig. 3. Due to the position of the pilot in front of the rotational axis, both yaw rotation cues as well as translational acceleration cues were present in the yaw tasks. The experiments were performed on the VMS with one to one motion simulation. - Figure 4. Simulator cockpity motion configurations in plan view. Schroeder distinguished four different motion feedback configurations, which have been applied in the experiments and the analysis, Fig. 4. These configurations are: 1 Translation and yaw acceleration feedback 2 Translation acceleration feedback 3 Yaw rotation feedback 4 No motion feedback When McRuer postulated his quasi-linear pilot model, he described the pilot s behavior with a linear model and added a noise signal as remnant n(t) to the output to describe the non-linear part of pilot s behavior. Where no information on remnant was available the linear pilot model was used without a remnant signal. This influenced the performance in minimizing the yaw error and the pedal control output in the analysis but hardly the model control behavior. The results of these experiments showed that translational acceleration did have a large significant contribution to pilot s control and performance while the yaw feedback did not contribute. Based on the results Schroeder concluded that yaw rotational platform motion is not adding value to helicopter flight simulation. However, the claim that yaw rotational platform motion does not provide a cue is not made. Analysis Initially the analysis was set up for the yaw capture task but once set up it was easy to extend the analysis for additional tasks to broaden the analysis. In addition to the yaw capture task, two continuous tracking tasks were analyzed. One was a target following task where the pilot forces the aircraft to follow the external Forcing function i(t), Fig. 1. The other was a disturbance rejection task where the aircraft is disturbed by an external disturbance w(t). Both external signals i(t) and w(t) were pseudo random signals based on a sum of sinusoids. There is a distinct difference between the influence of motion feedback in the target following task and the disturbance rejection task. The origin of this difference results from the different position where the external disturbance enters the control loop,. The closed loop structure used for the analysis is presented in Fig. 1 for the real aircraft. For the simulated aircraft, the closed loop was extended with the specific characteristics of the simulation (time delays, and motion system dynamics and washout) as described by Schroeder (1999), Fig. 5. In this particular case one to one motion was applied and washout was not taken into account in the analysis. 4

5 Figure 5. Closed loop structure for the simulated vehicle. From Schroeder s experimental results, no information on the crossover frequencies for the different experimental condition could be obtained. Therefore, the crossover frequencies were adjusted based on the vehicle dynamics and the impressions from the pilot s response in the yaw capture task. Results The results of the initial analysis confirmed the results of Schroeder s experiments. In the yaw capture task the translational acceleration improved the tracking performance and control behavior while yaw feedback hardly influenced the performance and control behavior. In the target following task the feedback of translational motion changed the control behavior while the performance hardly changed. The influence of yaw rotation feedback was marginal. However, in the disturbance rejection task the analysis predicted a strong influence of both yaw rotation as well as translational acceleration on both performance and control behavior. This latest finding was in contradiction with the findings of Schroeder who concluded that yaw rotation was not effective. After the workshop Peter Grant from the University of Toronto asked the analysis data to perform an experiment with the control tasks analyzed. After an extensive exchange of data about the simulator experiments with Schroeder he was able to perform the yaw capture task, the target following task and the disturbance rejections task on the UTIAS Flight Research simulator (CAE 300 series Steward motion system). This gave considerable concern due to differences in time delays, motion system dynamics and motion space. With great care the UTIAS simulator was configured as close as possible to the VMS for these experiments (Grant et al., 2005) but differences remained. Based on the results Grant concluded: The repetition of Schroeder s yaw capture study on the UTIAS facility did not lead to identical results. In particular, the current study found that yaw motion alone can be of benefit to pilot performance, although it appears that the additional benefit from yaw motion is quite small once translational motion is present. Translational motion was also demonstrated to improve performance, but unlike Schroeder s study, there was little additional benefit when yaw motion was present. The study demonstrates the difficulty in repeating experiments at different facilities. It is a large effort and there are often simulator differences that cannot be eliminated. 5

6 For a representative disturbance task, it appears that both yaw and translational motion can be of benefit to performance, although the benefit from translational motion is larger. For the tracking (target following) task there was a small but significant effect of translational motion and a small marginally significant effect of yaw motion on performance. Hosman s descriptive pilot model proved useful at determining the tasks that were studied in the simulator. Conclusions: The pilot model analysis performed based on the NASA Ames helicopter simulator experiment showed that a valuable extension of the experiment could be indicated. The results of the new experiment performed at UTIAS confirm the findings of the analysis about the positive contribution of yaw rotation in disturbance rejection tasks. 3. Pilot model based optimization of the cueing algorithm parameters Figure 6. Integrated motion-cueing system design process. During the Flight Simulation Conference The Next Decade in 2000 an integrated design of the cueing algorithm and the motion system was proposed (Advani and Hosman, 2000). The concept was to develop an optimal design of the total motion cueing system by optimizing the motion space of the 6 DOF motion system interactively with the cueing algorithm to find the best solution for a specific application, Fig. 6. The choice of motion system characteristics and cueing characteristics is based on the requirements of the simulator user. The major part of the motion system requirements, in particular the required motion space, system bandwidth, etc., are actually dependent upon the task to be simulated. The lack of motion-cueing criteria in the regulations, however, is the primary reason that there has been no real progress in the development of motion-cueing algorithms and the systematic design of simulator motion systems during the last decades. As a result, the final specifications are primarily based on empirical knowledge and experience, rather than on the actual training need. The integrated design is based on a sequential analysis, which leads to the optimization of the motioncueing algorithm and the geometry of the motion system. The authors developed a systematic means of specifying simulator motion-cueing systems, incorporating the motion drive algorithm and the 6

7 motion-base mechanism (Fig. 6). Several projects were performed to define the total motion cueing system for particular applications (Hosman and Advani, 2005a, 2005b). This paper concentrates in this section on the application of the pilot model for the analysis to optimize the cueing algorithm parameters. As shown in the previous section the pilot model, Fig. 2, can be adjusted to particular aircraft dynamics and a specific control task. After the model has been adjusted to normal pilot control behavior in real flight, the closed loop of Fig.1 is extended with the influence of the simulation process, Fig. 7. Disturbance w(t) Forcing function i(t) Outside Visual and instruments e(t) Pilot model u(t) Aircraft model y(t) Simulated aircraft Computer time delay Vestibular feedback Motion system Wash-out filters Visual feedback Figure 7. Pilot simulated-aircraft control loop. In this scheme, four elements are added to the pilot aircraft control loop: the host computer time delay, the wash-out filter, the motion system, and the visual systems and their dynamic characteristics and respective time delays. In the start of the analysis, the characteristics of the motion system are estimated. In later steps of this iterative process a better estimate of the motion system characteristics will be used. The analysis is performed separately for pitch-surge, roll-sway, yaw and heave. Due to the changes of the control loop as caused by the simulation process, the pilot will have to adapt his behavior to the simulated-aircraft control loop However, the goal of the washout 23 adjustment is to keep pilot s behavior 22 as close as possible to that in the 21 aircraft. Using the pilot model and the aircraft model as the basis, the washout 18 algorithm parameters are adjusted J 17 by hypothesizing that the cost function (eq. 1) used to adapt the pilot model to 14 the aircraft dynamics indeed describes the pilot s control strategy. Therefore, the washout parameters are adjusted Wnhp using the same cost function The example given below is based on Wnlp the cueing system for a Wright Flyer simulator. After the pilot model was Figure 8. Cost function J as a function of the bandwidth of the tilt coordination filter ω n (Lp) and rotation filter ω n (Hp) for the roll disturbance task

8 adapted to the aircraft dynamics, the pilot model parameters were fixed, and the pilot simulatedaircraft loop was established. The wash-out filter parameters were optimized for the simulation of the Wright Flyer with a lumped simulation time delay of 70 ms. The parameters of the cueing algorithm were adapted using the cost function J of Eq. (1). In Fig. 8, as an example, the cost function is shown with respect to the low pass Hp( ω ) Hp( ω) ω (rad/s) ω (rad/s) ϕ (degr) ϕ (degr) ω (rad/s) Pilot model ω (rad/s) Pilot model washout filter included Visual feedback, aircraft Visual and vestibular feedback, aircraft Visual and vestibular feedback, simulated aircraft, classical washout Visual and vestibular feedback, simulated aircraft, optimal filter Figure 9. Bode plot of the pilot model for the Wright Flyer with and without motion feedback, and for the simulated aircraft with the classical washout filter and the optimal filter. tilt-coordination and high pass pitch rotation filter bandwidth ω n (Lp) and ω n (Hp), respectively. In most cases, this cost function J can be described by a valley as a function of the break frequencies of the high-pass rotational filter and the low-pass tilt-coordination filter. After the wash-out parameters have been adjusted, the pilot model can be readapted to the simulated aircraft with the adjusted cueing algorithm, and subsequently, the pilot model adaptation to the real and the simulated aircraft can be compared. The left bode plot of Fig. 9 shows the pilot model transfer function for four conditions: 1. Real aircraft without motion 2. Real aircraft with motion feedback 3. Simulated aircraft with classical washout (CW2, Reid and Nahon, 1986) 8

9 4. Simulated aircraft with optimized washout. For the design of the motion cueing system of the proposed Wright Flyer simulator the actuators of a small commercial available motion system were used while the workspace of the system was reshaped by changing only the locations of the attachment points. In this way a one to one motion simulation was possible for the foreseen Wright Flyer flight maneuvers. Mostly, to keep the simulator motions within the motion space of the motion system the washout algorithm attenuates the simulator motions. This means that the motion feedback to the pilot in the simulator is also attenuated. To compensate for this attenuation and to make optimal use of the motion cues, the simulator pilot has to increase the weight of the sensory output of the vestibular system. To demonstrate this also the CW2 washout (Reid and Nahon, 1986) which has a gain of 0.6 was applied. The left bode plot of Fig. 9 shows the pilot models for the four different conditions mentioned above. The right bode plot presents the same bode plots of the pilot model but corrected for the cueing algorithm. This bode plot shows that the transfer function describing pilot s control behavior in the simulator with optimized washout closely corresponds with the control behavior in real flight. In 2006 the author proposed a motion cueing criterion covering the whole motion cueing system, from aircraft model output to the simulator motion (Advani, Hosman, 2006). During the AIAA MST Motion Workshop in 2006 the author invited research institutes to make the frequency response of the motion cueing of their research simulators available to support this development. The results were published in 2007 (Advani, Hosman, Potter,2007). In Fig.10 the results of an optimized motion cueing system from 2002 is plotted together with the results of the research simulators in Nichols plots for pitch and surge. The basic assumptions of the adjustments of these cueing systems have to be taken into account when comparing these plots into detail. Mind that the optimized motion cueing system plots are based on transfer functions while the frequency response of the research simulators was measured at a limited number of frequencies. Figure 9. Optimized motion cueing for pitch and surge compared with those of research simulators (2006) of UTIAS, TsAGI and JAXA. Conclusions: The results of the washout adjustment based on a pilot model show that positive results are obtained which provide a comprehensible solution. The big advantage is that the algorithm parameters are 9

10 adjusted without the dependence on the subjective judgment of pilots which often lead to contradictory solutions. 4. Pilot model application for the Balked Landing Study. Based on an initiative of the ICAO Instrument Flight Procedure Panel (IFPP, former OCP) pilot models are under development capable to control a transport aircraft model during a simulation through an approach to land, balked landing and subsequent climb out. The goal of this pilot model development is to apply these models in a Monte Carlo Simulation to determine the flight path statistics during the balked landing of the New Large Aircraft (A380 and B747-8) to verify the Obstacle Free Zone for ICAO Code F aircraft (wingspan > 65 m). This was considered necessary due to the fact that the original OFZ was based on measurements in the 60 th and considerable improvement of both flight control systems and landing aids have been achieved since. The so-called Balked Landing Study as initiated by the IFPP is strongly supported by the FAA (Monte Carlo simulations with ASAT) and the Boeing Aircraft Company by making the B aircraft model available and taking care of the integration of the pilot models with the hi-fidelity model. In addition, NLR/AMS Consult, QinetiQ, NASA Ames, Boeing Research and Development Europe and TsAGI cooperate in this international study. Two independent pilot models are under development; one at QinetiQ and one at NLR/AMS Consult. TsAGI will be responsible for the final validation of the pilot models. Boeing Research and Development Europe support the study by additional research where appropriate. The NLR/AMS Consult pilot model is based on the descriptive pilot model discussed already in Section 3. However, there were some important additions necessary to make the model suitable for the application. The original model was developed for skill based attitude control. During the approach to land followed by a balked landing the pilot task changes considerably and therefore the pilot model has to be adapted to these changes. To make that possible, the whole maneuver was split up into five segments: 1. Flight director approach down to an altitude no less than 200 ft. 2. Visual segment down to flare and de-crab initiation at ± 50 ft. 3. Flare and de-crab. 4. Pitch attitude rotation and re-crab after go-around initiation. 5. Flight director climb-out up to 400 ft. For each segment the pilot model is adjusted and extended for symmetric and asymmetric flight path control. In addition, a model for speed/thrust control for the flight director and visual segment is added. The real challenge of the study is that the pilot model in the final Monte Carlo simulations has to perform statistically equal to real flight performance. To support the pilot model development and validation, simulations of the balked landings have been performed and recorded in the NASA Ames 747 FFS and in the Boeing 747 fixed base engineering simulator. The 747 aircraft model is provided by Boeing in the control engineering EASY5 environment. In the following, some details of the model development necessary to obtain the required accuracy in tracking performance will be discussed. 10

11 Both pilot models describe pilot s skill-based control behavior. On top of these models a Procedure model was developed at the NLR to initiate all discrete procedural actions. This model is based on a statistical analysis of pilot s rule based behavior during the NASA Data trials. Below some aspects of the pilot model will be discussed. For a more extended description of the pilot model refer to Hosman et al. (2009). Pilot s non-linear behavior In the particular case of the application of the pilot model to the Balked Landing Study, it is required to extend the model to simulate the non-linear behaviour of the human operator. In real life, pilots' control behaviour will exhibit non linear behaviour due to inaccuracies and errors (McRuer, 1965). To take that into account, mostly a remnant signal n(t), a bandwidth-limited noise signal, is added to the model control output, as shown in Fig. 10 (Clement et al, 1968, Steurs et al, 2004). This provides a good solution for experimental control tasks. Figure 10. Pilot model remnant n(t) as proposed by McRuer(1965). However, in real flight a feature of pilot s behaviour in generating control outputs, and related to the non-linear characteristic, is that a pilot will often not react immediately to small deviations. With a closer look at the time histories of the NASA data files it turns out that the subject pilots during the FD segment, visual segment, and the climb out generate a pulse like control output when compensating for errors rather than a continuous control signal. This can be modeled by generating an output only if a certain threshold is exceeded. By positioning this threshold before the 2 nd order neuromuscular system, the step like changes in the control signal due to the threshold are smoothed, Fig. 11. By adding the remnant directly to the model output, it will be non-zero while the control signal may be zero. To avoid this, the remnant signal is multiplied with the control output. After the control signal passed the threshold it is multiplied with the noise signal. Thereafter the remnant is filtered and added to the control signal. In this way, the remnant is zero when the control signal is zero and remnant magnitude varies with the control output magnitude. The threshold value itself depends on pilot s control task, i.e. symmetric and asymmetric control and flight segment. 11

12 Figure 11. Pilot model with remnant incorporated. The visual segment The visual segment is the last part of the manual flown ILS approach. The pilot controls the aircraft primarily based on the outside visual scene. It starts at or above decision height ( 200 ft) when the aircraft approaches the runway and ends at the start of the flare and de-crab or the go-around. During the visual segment, the pilot primarily looks outside and has to interpret the outside visual scene to determine his position relative to the intended flight path corresponding with the extended runway centerline and the - 3 glide path. To correct for the lateral and vertical deviations to the intended approach path, the pilot has to close the inner attitude loops and the outer flight path and position loops, Figs. 12 Pilotmodel inner loop Von Karman gust Pilot lateral position uncertainty + e ref e y + Pilot gain ϕ + Pilot lateral ref Controller K χ eϕ Linear pilot model Remnant + + Vestibular feedback δ Aircraft model y Roll attitude feedback Track feedback Lateral localizer deviation feedback Figure 12. Lateral control during the visual segment. The pilot model parameters for the inner loop control are adjusted with an optimization procedure as discussed in Section 2. The bandwidth of the inner control loops depends on the aircraft characteristics, i.e. the short period mode and the roll mode. The outer loop pilot gains are adjusted according to Hess (1987). The bandwidth as expressed by the crossover frequency of the outer loops is a fraction of the bandwidth of the inner loop. 12

13 In addition to the remnant, describing pilot s inaccuracy and non-linearity in generating his control output, pilot s inaccuracy in perceiving the aircraft s position relative to the intended approach path has to be added to ensure a realistic performance of the pilot model for the present application. Based on the work of McRuer (1965, 1967) on pilot s control behavior, Clement et al (1968) on pilot control during the manual ILS approach, and by Wewerinke (1978, and 1980) on perception accuracy during the visual segment, pilot s perception uncertainty of the aircraft position can be modeled. This is modeled by inserting a slowly varying random bias (pilot position uncertainty), Figs12, with a magnitude dependent on the distance to the touch down aiming point, Fig. 13. Figure 13. The standard deviation of the perception uncertainty of height Δh and lateral deviation Δy as a function of the distance to the touch down aiming point. De-crab and re-crab after the go-around initiation In the sequence of the approach/balked landing so far, the pilot model has performed the approach to land by elevator and aileron control with zero sideslip. In case of a crosswind, the aircraft heading deviates from the track angle by the drift or crab angle. Before touchdown, the pilot may de-crab the aircraft by correcting the aircraft heading with rudder input to correspond with the track angle and correcting for the resulting drift by rolling the aircraft into the crosswind. This has also to be accomplished by the pilot model. Figure.14. Lateral control during the visual segment extended with rudder control for the de-crab. 13

14 When initiating the de-crab with the sideslip technique, a rudder input is generated based on the side slip command while the drift is compensated by maintaining the required track by aileron input. In Fig. 14 a yaw control loop is added to the lateral control, Fig. 12, during the visual segment. The inner yaw loop model is activated when the de-crab is started and the pilot inner loop yaw model provides the required rudder input. The pilot inner loop yaw model is basically the descriptive pilot model as presented in Fig. 2. Track control by aileron input is too slow to compensate fast enough for the disturbance of the track angle resulting from the rudder input. The skilled pilot is aware of that effect and corrects proactively. Therefore, the rudder input is fed back to the pilot inner loop roll model to generate a compensatory aileron control input. The procedural model initiates the re-crab after the go-around is initiated by reducing the commanded side slip. For those cases where the go-around is initiated before the de-crab is started, no de-crab will be performed. Pilot model parameters are adjusted in three steps. First the inner roll attitude loop and second the lateral control loop are adjusted with the aircraft model for 30 ft for the flare and de-crab. Thereafter the yaw control loop is adjusted. Results So far, the NLR pilot model structure has been determined for longitudinal, lateral and speed/thrust control for the five segments of the balked landing maneuver. When it became clear that TsAGI will use the completed EASY5 model for the first step of the pilot model validation, the aircraft model in EASY5 was extended with the undercarriage, ground effect and a more refined engine model. In addition, it was decided to develop a Procedural model (van der Geest, 2009) describing pilot s discrete procedural actions during the balked landing maneuver. Presently, the final pilot model and the Procedural model are implemented in EASY5 and when integrated ready to be tested. To first test the integrated pilot model and procedural model, the pilot model was adjusted to one configuration of the NASA Data Collection. So far, performance data have been derived for the off line simulations in Matlab/Simulink in the subsequent segments and compared with the performance data from the NASA Data Collection as far as a clear vertical or lateral reference is available. The pilot model data are obtained from 10 realizations of the simulated atmospheric disturbance during the NASA Data Collection while the NASA data are valid for the particular flight segment and based on 13 pilot subjects and 4 replications. The pilot model parameters were varied to simulate the inter and intra pilot variability. The performance data are presented in Fig. 15. The final pilot model will first be adjusted to the 12 Configurations of the NASA Data Collection for validation of the pilot model performance. When validated, the pilot model will be adjusted to 36 aircraft configurations (aircraft weight, cg, flap setting, and indicated airspeed) for application in the Monte Carlo simulations. So far the pilot model development is completed and fine tuning of the model parameters to assure that the pilot model tracking performance will match the FFS simulator data will be performed this year. The question whether manual flown simulated balked landings have statistically the same tracking performance as manual flown balked landings in the day to day flight operation has still to be answered. Since a full pilot model for the balked landing has been developed an analysis with the 14

15 simulation of the B747 FFS will be performed to evaluate the influence of the simulation process on pilot s tracking performance. Figure 15. Off line pilot model performance compared with pilot subject performance during the NASA Data Collection. Pilot model data based on 10 runs, NASA Data based on 13 subjects and 4 replications. 5. Discussion and Conclusions In this paper three pilot model applications with respect to simulation have been discussed: 1. The use of pilot models for pre-experimental analysis 2. Pilot model based optimization of the cueing algorithm parameters 3. Pilot model application for the Balked Landing Study The results of these applications demonstrate that pilot model development has reached a full grown status and that pilot models are ready to be used for a wide range of applications. Where the first two applications were directed to analysis of pilot s control behavior of inner loop attitude control, the third application is directed to both pilot s control behavior and tracking performance in a task described with a set of nested control loops. This extends the possible range of applications. Although flight simulation developed enormously over the last 40 years, there are still may questions about the influence of the simulation process on pilot s control behavior and performance. Further development and improvement of simulation can be supported by the use of pilot model analysis. 15

16 Among others, it may support improvement of motion cueing and the development of a motion cueing criterion. References: Advani, S.A., and R.J.A.W. Hosman (2000). Integrated Motion Cueing Algorithm and Motion-Base Design for Flight Simulation. Proceeding of the conference on: Flight Simulation The Next Decade. Royal Aeronautical Society, London, May, Advani, S.K., and R.J.A.W. Hosman (2006). Revising Civil Simulator Standards An Opportunity for Technological Pull. AIAA Modeling and Simulation Technologies Conference and Exhibit August 2006, Keystone, Colorado Advani, S.K., R.J.A.W. Hosman, and M. Potter (2007). Objective motion fidelity qualification in flight training simulators. AIAA Modeling and Simulation Technologies Conference and Exhibit August 2006, Hilton Head, South Carolina Clement, W.F., H.R. Jex, and D. Graham, Application of system analysis theory for manual control displays to aircraft instrument landing, Forth Annual NASA University Conference on Manual Control, NASA SP-192, Geest, P van der, and M. Trujillo (2009). Development of a Procedural Pilot Model for Event Sequencing During the Manual Balked Landing Maneuver. AIAA Modeling and Simulation Technologies Conference. Chicago, August 10-13, Grant, P.R., B. Yam, R. Hosman, and J.A. Schroeder (2005). The effect of simulator motion on pilot s control behavior for helicopter yaw control tasks. AIAA Modeling and Simulation Technologies Conference. San Francisco, CA. August AIAA Hess, R.A. (1980), Structural model of the adaptive human pilot. Journal of Guidance and Control, Vol. 3no. 5, pp , 1980 Hess, R.A. (1987). Feedback Control Systems. In: G. Salvandy ed., Handbook of Human Factors. Wiley Interscience Publication. John Wiley & Sons, New York. Hosman, Ruud (1996). Pilot s perception and control of aircraft motions. Ph. D. Thesis. Delft University of Technology. Delftse Universitaire Pers, Delft, the Netherlands. Hosman, R.J.A.W., Analysis of the Yaw Experiment, Unpublished AIAA MST Motion Workshop, Austin, Texas, Hosman, R, P.R. Grant, J.A. Schroeder (2005). Pre and post pilot model analysis compared to experimental simulator results. AIAA Modeling and Simulation Technologies Conference. San Francisco, CA. August AIAA Hosman, R.J.A.W., S.K. Advani and N. Haeck (2005a). Integrated design of the motion cueing system for a Wright Flyer Simulator. AIAA Journal of Guidance, Control and Dynamics. Vol 28, nr. 1, pp Hosman, R.J.A.W., S.K. Advani and N. Haeck (2005b). Integrated desing of flight simulator motion cueing systems. RAeS, The Aeronautical Journal. February

17 Hosman, R.J.A.W., P.van der Geest and J. van der Zee (2009). Development of a Pilot Model for the Balked Landing Maneuver. AIAA Modeling and Simulation Technologies Conference. Chicago, August 10-13, Kleinman, D, L., and S. Baron (1971b). Manned Vehicle Analysis by Means of Modern Control Theory. NASA CR Levison, W.H., and A.M. Junker, A model for pilot's use of motion cues in roll-axis tracking tasks, Bolt, Beranek and Newman, Inc., Report No. 3528, Cambridge, MA, McRuer, D.T., D.Graham, E.S. Krendel, and W. Reisener (1965). Human Pilot Dynamics in Compensatory Systems. Theory, Models, and Experiments with Controlled Element and Forcing Function Variations. AFFDL-TR-65-15, Wright Patterson AFB, OHIO. McRuer, D.T., and H.R. Jex (1967). A Review of Quasi-Linear Pilot Models. IEEE Transactions on Human Factors in Electronics. Vol. HFE-8, no 3, September Reid, L.D. and M.A Nahon (1986). Flight Simulation Motion-Base Drive Algorithms: Part 2. Selecting the system parameters. University of Toronto, Institute for Aerospace Studies. UTIAS Report No 307. Schroeder, J.A.(1999). Helicopter Flight Simulation Motion Platform Requirements. NASA Ames Research Center. NASA TP Stapleford, R.L., R.A. Peters, and F.R. Alex, Experiments and a model for pilot dynamics with visual and motion inputs, NASA CR-1325, Steurs, M., M. Mulder, R. van Paassen, "A cybernetic approach to assess flight simulator fidelity", AIAA Modeling and Simulation Technologies Conference, Providence, RI. August AIAA Wewerinke, P.H. (1978). Visual scene perception process involved in the manual approach. National Aerospace Laboratory (NLR), Amsterdam. NLR TR U. Wewerinke, P.H. (1980). The effect of visual information on the manual approach and landing. National Aerospace Laboratory (NLR), Amsterdam. NLR TR U. 17