Zusammenfassung
Flight simulators provide an effective, efficient, and safe environment for practising flight-critical manoeuvres without requiring a real aircraft. Most simulators are equipped with a Stewart-type motion system, which consists of six linear actuators in a hexapod configuration. The argument for use of motion systems in simulators is derived from the presence of motion cues during flight. It is hypothesised that if pilots would train in a fixed-base simulator, they would adapt their behaviour and that this would result in incorrect control behaviour when transferred to the aircraft. Similarly, if pilots would train without simulator motion, the presence of motion in flight could disorient the pilot which could have a detrimental effect on performance. Finally, pilots themselves have a strong preference for vestibular motion cues to be present in flight simulators. Therefore, flight simulator motion systems are used to reproduce aircraft motion experienced in flight as faithfully as possible, and to provide the pilot with the most realistic training environment. Flight simulator regulators also allow the use of low-cost motion systems with reduced magnitude motion cues compared to full flight simulators for certain non-type specific training tasks. The limited characteristics of these motion systems, such as shorter actuators, lower bandwidth, and lower smoothness, are hypothesised to have an effect on pilot control behaviour in the simulator. Instead of relying on standard-practise subjective pilot ratings to determine these effects, it would be best to consider human perception and control processes at a skill-based level as a measure for the degree to which a simulator affects pilot perceptual-motor and cognitive behaviour for a given task and environment. Skill-based behaviour represents the lowest level of human cognitive behaviour and involves elementary human information processing and basic control tasks. Investigating this level of human behaviour provides an objective means to assess perception and control behaviour in a simulator environment. Skill-based behaviour can be assessed in simulator trials by taking a cybernetic approach, in which a mathematical model is fit to the measured response of a pilot and changes in the identified parameters serve as a measure for changes in human behaviour. The contribution of visual and vestibular information to control can be measured by performing closed-loop control tasks in which a pilot tracks a target, while at the same time rejecting a disturbance. Observed changes in performance can now be correlated with changes in identified control behaviour, and related to changes in experimental conditions. The goal of this thesis was to apply a cybernetic approach to investigate the influence of limited motion system characteristics of low-cost simulators on perception and control behaviour of pilots. Simulators with high-fidelity motion systems were used as a comparison. An initial motivation was the inconclusive evidence provided by previous studies on the influence of simulator motion, even though many experimental evaluations have been performed. A key reason for the lack of consensus is the limited understanding of human perception and control processes. A multi-modal cybernetic approach can provide a more detailed view by separating the contribution of individual perception channels. A second motivation was that it is unclear how human behaviour in the simulator is affected by limited motion system characteristics of low-cost motion systems. Two objectives were formulated towards the goal of this thesis: 1) assess the motion system characteristics that play a role in pilot perception and control behaviour, and 2) determine the influence of these characteristics on pilot control behaviour in experimental evaluations. By contrasting the limited characteristics of a low-cost motion simulator to those of a high-end simulator, it is possible to specify the properties of motion systems that are most important to human control behaviour. After modelling the properties of a low-cost motion platform and simulating that model on the high-end platform, the limiting motion system characteristics can be varied systematically to represent either simulator, or any 'virtual' simulator in between. The cybernetic approach can then be used to identify pilot control behaviour, and adaptation of pilot control strategies can be related to changes in the motion cues that are available during active control tasks in the simulator. To achieve the first objective, two research simulators were used to investigate the basic properties of simulator motion systems: 1) the MPI Stewart platform, a mid-size electric simulator with restrictive characteristics, and 2) the SIMONA Research Simulator (SRS), a larger hydraulic motion simulator with well-known properties. The characteristics of the MPI Stewart platform were determined using a standardised approach, in which the measured output signal from an Inertial Measurement Unit (IMU) was partitioned into several components in the frequency domain such that the various characteristics of the motion platform could be determined. These included the describing function, low and high frequency non-linearities, acceleration noise, and roughness. The primary finding from these measurements concerned the platform describing function, which was dominated by the standard platform filters implemented by the manufacturer. Outside the 1 Hz bandwidth of the platform filters, the signal-to-noise ratios were very low. Furthermore, the first-order lag constant from dynamic threshold measurements was relatively high, which meant that the platform response to an acceleration step input of 0.1 m/s^2 was slow and only reached 63% after approximately 300 ms. Initially, a relatively high fixed time delay of 100 ms was found between sending a motion command to the platform and measuring its response. The measurements revealed that this was related to the software framework used for driving the simulator, which was subsequently updated. This resulted in a much lower time delay of 35 ms. Based on these performance measurements, a model was developed for the main characteristics of the MPI Stewart platform: its dynamic range based on the platform filters, the measured time delay, and characteristics of the motion noise (or smoothness). After baseline response measurements were performed on the SRS, the model of the MPI Stewart platform was implemented and validated with describing function measurements. The baseline measurements on the SRS showed a dynamic response with a bandwidth higher than 10 Hz and a time delay of 25 ms. Measurements during simulation of the MPI Stewart platform model showed that the SRS could replicate the model response and time delay characteristics, and that the motion noise could be reproduced as well. Thus, the implementation of the total model of the MPI Stewart platform on the SRS was validated and systematic changes could be made to motion system dynamics, time delays, and motion noise characteristics to study their effect on human control behaviour. These findings achieved the first objective of this thesis. The second objective was addressed using a two-step approach. The first step consisted of developing a novel parametric technique for identification of human control behaviour and comparing it to an established spectral method using Fourier Coefficients. It was shown that the parametric method was able to reduce the variances in the estimates by assuming a pilot model structure and by incorporating the pilot remnant. Furthermore, the analytical calculations for bias and variance in both methods were validated with the use of 10,000 closed-loop simulations, and the methods were successfully applied to experimental data of closed-loop multi-channel control tasks. In the second step, it was investigated how the simulator motion system characteristics affected pilot control behaviour, by simulating the model of the MPI Stewart platform on the SRS. The model characteristics were varied systematically in a closed-loop control experiment with simultaneous target and disturbance inputs, such that pilot control behaviour could be estimated for visual and vestibular perceptual channels. Participants performed a pitch tracking task, using a simplified model of the pitch attitude dynamics of a Cessna Citation I. At the same time they rejected a disturbance on their control input. Simulator motion cues were presented in pitch and heave. However, only vertical motion due to rotations around the centre of gravity were considered in this experiment, and the influence of centre of gravity heave was not taken into account. It was shown that the 1 Hz platform filter of the MPI Stewart platform had the largest experimental effect. The bandwidth of the motion system response was limited drastically compared to the baseline SRS response. Participants could not reduce tracking errors effectively, and barely used the motion cues at all in conditions with a limited motion system bandwidth. Instead, participants relied on visual cues to generate lead in their control behaviour necessary for the control task. The experimental evaluation did not show an influence of the difference in simulator time delays (35 ms versus 25 ms) on pilot control behaviour. Similarly, the simulator motion noise characteristics did not have an effect. The disturbances in motion cues due to these characteristics were not large enough to obscure motion information that was relevant to the control task, as the difference in time delay between the MPI Stewart platform and the SRS was only 10 ms and the motion cues due to the motion noise characteristics were small. Therefore, these motion system characteristics did not impair the ability of pilots to generate lead information from the motion cues for the task used in this experiment. However, these motion system characteristics could have a different effect in other experimental tasks, such as measurements on pilot motion thresholds. The second objective of this thesis was fulfilled by determining the influence of motion system characteristics of two research simulators on pilot performance and control behaviour. Future research should focus on applying the cybernetic approach to other types of motion systems. Full flight simulators with electric actuators are a prime candidate for this approach as they are replacing hydraulically driven simulators, and specifications about their motion systems are rarely published. Furthermore, flight simulators are mainly used for pilot training. Simulator motion rarely shows an effect in studies on transfer of training from simulator to aircraft, whereas it can have a pronounced effect on pilot control behaviour as has been shown in this thesis. Efforts to bridge the gap between these research fields should investigate requirements for simulator motion in pilot training, for motion system tuning, and for experimental control tasks. A related research question exists in understanding the influence of simulator motion in more ecologically valid piloting tasks. Higher-level piloting tasks could be investigated by extending the cybernetic approach to more cognitive aspects of human behaviour. Additionally, more basic research is required for looking into the different components that contribute to forming a percept of motion. For instance, the influence of proprioception and somatosensory feedback is not well understood. The approach used in this thesis provided valuable insight into changes in pilot response dynamics that form the basis of observed changes in performance. The results demonstrated that simulator motion cues must be considered carefully in piloted control tasks in simulators and that measured results depend on simulator characteristics as pilots adapt their control behaviour to the available cues.
