Peter Zaal

Aerodynamic Hinge Moment Coefficient Estimation Using Automatic Fly-by-Wire Control Inputs

This paper describes the modeling and parameter estimation of the aileron and elevator flight control system of TU Delft’s Cessna Citation II laboratory aircraft. The flight test data originate from maneuvers performed autonomously with a custom-designed experimental fly-by-wire system. The identification of the aerodynamic hinge moment coefficients will be of special interest, as these hinge moments greatly affect the in-flight performance of the flight control system. First, the elevator and aileron flight control system models will be presented, introducing the main parameters that need to be determined. Most of the parameters reflect the mechanical properties and can be obtained through some cleverly-designed ground tests, which are discussed next. The hinge moment coefficients can only be determined through flight tests. The paper continues with a description of the optimal input signals used to generate flight data for the parameter estimation procedure. The flight test setup will be introduced briefly, after which the results of the hinge moment coefficient parameter estimation are summarized. Finally, the validity of the resulting flight control system models for elevator and aileron are shown.

Effects of Motion on Pilot Behavior in Target, Disturbance and Combined Tracking Tasks

In vehicle control, pilots merge the input from multiple sensory channels to produce the control input to the vehicle. In order to separate different channels from a multi-channel pilot model with the current identification techniques, multiple forcing functions need to be introduced in a simulator experiment. Then, it is assumed that “classical” target and disturbance tasks can be replicated by properly downsizing the power of the disturbance and target forcing function signals, respectively. It remains unclear, however, whether pilots adapt their control strategy in these combined tasks, in particular in the presence of motion. In order to analyze pilot behavioral adaptation, a multi-channel Optimal Control Model is developed and a simulator experiment is conducted. A range of combinations of forcing functions is tested in a simulator experiment. However, no apparent change of pilot behavior was found.

A Cybernetic Approach to Assess Simulator Fidelity: Looking back and looking forward

This paper describes some of the main results of the VIDI project (2006-2012) which aimed to assess flight simulator fidelity through a model-based, cybernetic approach. In a number of experiments conducted in the SIMONA research simulator and in real-flight, we determined multimodal visual-vestibular pilot control models. Taking the in-flight pilot models as a baseline we were able to obtain an objective quantification of behavioral discrepancies measured in the simulator, for a large range of simulator motion cueing settings. Even though a perfect match is not evident from our experiments, it is found that simulator motion cueing best approximating the motion cues pilots are subjected to in real flight also induces tracking behavior that best matches in-flight measurements. Future work should aim at developing pilot models for more realistic flight tasks, requiring a leap forward in cybernetics theory, and focus on quantifying the effectiveness of simulator training to further optimize training programs and apparatus.

Time Domain Pilot Model Identification Using Maximum Likelihood Estimation

Currently, multi-channel pilot models parameter estimation is done using a two-step frequency domain technique—identifying a non-parametric frequency response and fitting a parametric model to it. A time domain identification method would only require one step— directly fitting a parametric pilot model to the time domain data. Time domain identification has additional advantages in that the forcing functions used do not have specific limitations and multi-channel identification can be accomplished with only one forcing function. This paper displays the results of single- and multi-channel pilot model identifications done using the MATLAB MMLE3 toolbox. Identification was performed first on Simulink simulations and then on actual data collected using the SIMONA Research Simulator at the Delft University of Technology. 6 . A time domain identification method only requires one step—directly fitting a parametric pilot model to the time domain data. Such an identification method may have several advantages over frequency domain methods. The forcing functions used do not need to be sums of sine waves, as the Fourier Coefficient method requires. For example, the transient response to a step input may be used to perform the identification in the time domain. Furthermore, it may be possible to identify a multi-channel pilot model with just a target signal or just a disturbance signal, but not both, as frequency domain methods require. This study is concerned with testing a method for performing multi-channel pilot model identification in the time domain. The method looked at was the Maximum Likelihood Estimation method. This method was originally developed to identify aerodynamic coefficients of aircraft, but can be applied to a pilot model. The pilot model that is used is the Van der Vaart model, which uses central visual and motion perception paths with constant gains and time delays, along with neuromuscular dynamics, to represent the pilot control behavior. A brief explanation of pilot model identification is given in Section II, followed by a discussion about the MLE method in Section III. The method was initially used to identify a pilot model simulation done in MATLAB Simulink. Information about the simulation setup is presented in Section IV. The identification results are presented in Section V including the methods sensitivity to initial parameter guesses, as well as robustness to remnant noise gain. In addition to the simulations, experimental data collected using the SIMONA Research Simulator (SRS) were also analyzed. Here the method was used to make both single- and multi-channel identifications. A discussion of the results is presented in Section IV.

Identification of Time-Varying Pilot Control Behavior in Multi-Axis Control Tasks

Recent developments in fly-by-wire control architectures for rotorcraft have introduced new interest in the identification of time-varying pilot control behavior in multi-axis control tasks. In this paper a maximum likelihood estimation method is used to estimate the parameters of a pilot model with time-dependent sigmoid functions to characterize timevarying human control behavior. An experiment was performed by 9 general aviation pilots who had to perform a simultaneous roll and pitch control task with time-varying aircraft dynamics. In 8 different conditions, the axis containing the time-varying dynamics and the growth factor of the dynamics were varied, allowing for an analysis of the performance of the estimation method when estimating time-dependent parameter functions. In addition, a detailed analysis of pilots’ adaptation to the time-varying aircraft dynamics in both the roll and pitch axes could be performed. Pilot control behavior in both axes was significantly affected by the time-varying aircraft dynamics in roll and pitch, and by the growth factor. The main effect was found in the axis that contained the time-varying dynamics. However, pilot control behavior also changed over time in the axis not containing the time-varying aircraft dynamics. This indicates that some cross coupling exists in the perception and control processes between the roll and pitch axes.

Pilot's Use of Heave Cues in a Pitch Control Task

This paper reports on a pitch control experiment performed in the UTIAS Flight Research Simulator (FRS), investigating the effects of pitch and heave motion on pilot performance and control effort in a pitch attitude control task. Few pitch attitude control experiments have been reported that investigate the influence of heave and pitch motion on pitch control. Moreover, experimental results, as reported in literature, are in disagreement about the influence of heave motion on pitch control. In this paper, the preliminary results of the experiment are presented. Two types of heave motion cues are studied. The first is a linear function of pitch rate, representing heave acceleration as a result of aircraft lift, the second is a linear function of pitch acceleration, representing heave acceleration as a result of the rotating aircraft. A combined target tracking and disturbance rejection task was performed with a compensatory display. The dependent measure analysis shows a significant increase in performance when pitch motion is added. An improvement in performance is visible for heave motion, though it is not significant. No significant differences are found for the two types of heave motion. Further in-depth analysis is currently conducted, including pilot describing function analysis by identification, and pilot model analysis.

Dual Extended Kalman Filter for the Identification of Time-Varying Human Manual Control Behavior

A Dual Extended Kalman Filter was implemented for the identification of time-varying human manual control behavior. Two filters that run concurrently were used, a state filter that estimates the equalization dynamics, and a parameter filter that estimates the neuromuscular parameters and time delay. Time-varying parameters were modeled as a random walk. The filter successfully estimated time-varying human control behavior in both simulated and experimental data. Simple guidelines are proposed for the tuning of the process and measurement covariance matrices and the initial parameter estimates. The tuning was performed on simulation data, and when applied on experimental data, only an increase in measurement process noise power was required in order for the filter to converge and estimate all parameters. A sensitivity analysis to initial parameter estimates showed that the filter is more sensitive to poor initial choices of neuromuscular parameters than equalization parameters, and bad choices for initial parameters can result in divergence, slow convergence, or parameter estimates that do not have a real physical interpretation. The promising results when applied to experimental data, together with its simple tuning and low dimension of the state-space, make the use of the Dual Extended Kalman Filter a viable option for identifying time-varying human control parameters in manual tracking tasks, which could be used in real-time human state monitoring and adaptive human-vehicle haptic interfaces.

Manual Control Cybernetics: State-of-the-Art and Current Trends

Manual control cybernetics aims to understand and describe how humans control vehicles and devices using mathematical models of human control dynamics. This “cybernetic approach” enables objective and quantitative comparisons of human behavior, and allows a systematic optimization of human control interfaces and training associated with manual control. Current cybernetics theory is primarily based on technology and analysis methods formalized in the 1960s and has shown to be limited in its capability to capture the full breadth of human cognition and control. This paper reviews the current state-of-the-art in our knowledge of human manual control, points out the main fundamental limitations in cybernetics, and proposes a possible roadmap to advance the theory and its applications. Central in this roadmap will be a shift from the current linear time-invariant modeling approach that is only truly valid for human behavior under tightly controlled and stationary conditions, to methods that facilitate the analysis of adaptive, and possibly time-varying, human behavior in realistic control tasks. Examples of key current developments in the field of cybernetics—human use of preview, predictable discrete maneuvering, skill acquisition and training, time-varying human modeling, and neuromuscular system modeling—that contribute to this shift are presented in this paper. The new foundations for cybernetics that will emerge from these efforts will impact all domains that involve humans in manual and semiautomatic control.

Between-Subject Variability in Transfer-of-Training of Skill-Based Manual Control Behavior

This paper describes a new approach for analyzing training effectiveness in transfer-of-training experiments, by considering the between-subject variability of post-transfer changes in task performance and control activity of individual trained pilots. First, exponential learning curve models were fit on experimental data of individual pilots. Second, curve parameters were used to analyze the immediate changes in task performance and control gain following transfer, and the correlation between immediate changes in task performance and continued learning rate after transfer. Data from two experiments with different experimental designs were compared using the new approach. The method revealed similar post-transfer effects in the immediate changes in task performance and control gain following transfer between the two experiments when pilots trained without motion. However, differences in post-transfer effects were found when comparing the correlations between the immediate change in task performance and learning rate. In addition, differences were found between participant groups training with different levels of flight simulator motion fidelity.