International Journal of Aerospace Engineering

International Journal of Aerospace Engineering / 2017 / Article

Research Article | Open Access

Volume 2017 |Article ID 7849461 |

Massimo Gennaretti, Federico Porcacchia, Simone Migliore, Jacopo Serafini, "Assessment of Helicopter Pilot-in-the-Loop Models", International Journal of Aerospace Engineering, vol. 2017, Article ID 7849461, 14 pages, 2017.

Assessment of Helicopter Pilot-in-the-Loop Models

Academic Editor: Christopher J. Damaren
Received29 Dec 2016
Revised27 Feb 2017
Accepted08 Mar 2017
Published27 Apr 2017


The aim of this paper is the evaluation of several pilot models found in the literature, suited for helicopter pilot-assisted and pilot-induced oscillations analyses. Three main topics are discussed: (i) sensitivity of rotorcraft-pilot couplings simulations on the application of the different pilot models available in the literature; (ii) effect of vehicle modeling on active pilot modeling; (iii) effects of interactions between active and passive pilot models. The focus is on hovering flight, where a specific adverse rotorcraft-pilot coupling phenomenon, the vertical bounce, may occur. Pilot models are coupled with a comprehensive aeroservoelastic model of a mid-weight helicopter. The numerical investigations are performed in frequency domain, in terms of eigenanalysis and frequency response analysis.

1. Introduction

Machine-pilot couplings are a wide class of phenomena potentially affecting every sort of vehicle, with consequences that may vary from discomfort to severe accidents. While being well documented in aeronautics (where they are usually called aircraft-pilot couplings, APC) since the beginning of the twentieth century [1], they were often ascribed to pilot faults, due to both lack of preparation or incorrect evaluation. Consequently, they received limited attention, mostly during training, when the pilot was instructed to perform particular actions, like to release controls, if an anomalous aircraft behavior was arising. On the contrary, APCs have been ignored during design process for decades. The result was a number of accidents (even recently, like that on the Japan Airlines flight JAL 706 [2, 3], in 1997) that has been reported as inability of the pilot to deal with unexpected aircraft oscillations (in that case arising when he took controls without disengaging the autopilot).

However, in the last decades, the scientific and technical communities have become more conscious about the importance of considering pilot and aircraft as a whole, in order to avoid such, potentially catastrophic, phenomena. In that sense, the role of the pilot in their insurgency has been partially reduced, while more attention has been devoted to the assessment of the conditions under which the aircraft-pilot coupling may yield undesired aircraft response. APC events have been defined alternatively as “involuntary trajectories and flight behaviors originating from an anomalous interaction between the pilot and the aircraft” or as “pilot-in-the-loop instabilities” [4], in both cases remarking the need of an active, though not necessarily voluntary, pilot participation. The attention devoted to APCs has allowed the identification of undiscovered potential instabilities in some cases related to dynamics traditionally considered stable. As an example, it has been observed that the short-period mode damping may consistently drop in presence of pilot in the loop [5].

Additional challenges in the analysis of human-vehicle interaction arise nowadays, mostly due to the increased presence of flight control systems in modern aircraft. Acting in parallel or in series with the pilot, they may lose effectiveness or become source of instabilities if their interaction with him is not carefully assessed during the design phase.

Although cases of rotorcraft-pilot couplings (RPCs) are reported since World War II [6], studies on RPC have lagged about thirty years with respect to fixed-wing counterpart, the pioneering works of Mayo [7] and Pharam [8, 9] being dated back to the last decade of the twentieth century. This fact, combined with the greater proneness of rotorcraft to instabilities, results in a wider area of relatively unexplored research. In the last years, the EC-funded FP7 research project ARISTOTEL has been mainly focused on the analysis of RPCs insurgency and on means for their prevention. In that activity, the authors have been mostly involved in the assessment of aeroelastic triggers of RPCs, with particular emphasis on main rotor-airframe aeroelastic coupling [1014].

Nevertheless, pilot modeling is still an open issue, usually addressed by separating active (voluntary, behavioral, and low-frequency) actions [1517] from passive (involuntary, biodynamic, and mid-frequency) actions [79, 1821], following the classification developed for fixed-wing aircraft. As a consequence, active pilot operations are more related to flight dynamics response phenomena, while passive pilot actions couple with aeroelastic responses.

The objective of the present work is the assessment of the following issues regarding pilot-in-the-loop rotorcraft response and stability analyses: (i) RPCs simulations sensitivity on the application of the different pilot models available in the literature; (ii) effect of vehicle modeling on active pilot modeling; (iii) effects of interactions between active-passive pilot models.

2. Rotorcraft-Pilot Coupling Classification

The RPC study requires competence in the modeling of several subsystems. Indeed, several different helicopter dynamics phenomena occur in the frequency range where RPCs events take place (as shown in Figure 1 for a medium-weight vehicle). These, combined with voluntary and involuntary pilot actions, leads to the definition of two main different kinds of RPCs: the so-called pilot-induced oscillations (or rigid-body RPCs), which are mostly related to active piloting and rigid-body dynamics, and the pilot-assisted oscillations (or aeroelastic RPCs) characterized by the involuntary feedback on controls due to cabin vibrations and mostly involving aeroelasticity and structural dynamics. The former involve events in the frequency range from to Hz, while the latter occur in the frequency range from to Hz (above, both pilot biodynamics and controls actuator responses become negligible).

It is worth noting that this separation is quite arbitrary, since an overlap between active and passive pilot behavior exists. Moreover, unlike in the fixed-wing case, aeroelastic effects usually play a crucial role in rotorcraft flight dynamics, primarily due to the low-frequency main rotor deformation. However, the Hz boundary is commonly accepted as valid from both the aeromechanics and the biodynamics point of view.

Another commonly used classification derived from fixed-wing field is that discerning the characteristic of the dynamics driving a RPC phenomenon and its trigger. Category I RPCs are caused by linear dynamics, while strong nonlinearities (e.g., control saturations) determine categories II and III (in the latter, an external trigger like malfunctioning, gust, or pilot task is required). Finally, an informal category IV often overlapping with PAOs is associated with RPCs characterized by significant elastic deformations. While cat. I phenomena are usually investigated through eigenanalysis, cat. II and III require specific techniques (typically in the time domain) [4]. Cat. IV RPCs may be approached in different ways depending on the characteristics of the system: thus far, most of the analyses have been limited to linearizable systems, thus exploiting eigenanalysis (see, e.g., the investigation on vertical bouncing described in [11]).

Control designers have identified a number of retrofit solutions to alleviate RPCs, usually consisting of the introduction of small friction on controls. At the same time, the pilot community has identified piloting procedures to deal with specific RPCs. In the case of vertical bouncing, an instable loop involving collective control, main rotor coning, and vertical motion of pilot seat (see Figure 2), the procedure simply consists of loosening the grasp of the hand on the collective lever, weakening the feedback. It is however evident that the capability to predict potential adverse couplings between pilot and aircraft since the design phase would represent a great advantage in the production of safer helicopters.

3. Pilot Modeling

A schematic representation of the pilot-in-the-loop system is shown in Figure 3. The pilot action is modeled as the combination of active and passive behavior, both given as feedback to the machine response. The commands exerted by the pilot (both voluntarily and involuntarily) are processed by FCSs (if present) and executed by actuators. These produce loads and consequently vibrations and rigid-body motion which, in turn, force pilot dynamics. Note that the inputs to the whole system are the specific mission/task to be performed and potential external disturbances.

In the following, a review of the pilot models used in this work is outlined.

3.1. Passive Pilot Modeling

Involuntary pilot actions on controls are determined by the complex dynamics of pilot body (including bones, muscles, and articulations) in response to the vibration transmitted from seat and controls. The most common approach to introduce passive pilot modeling in the loop is the use of transfer functions (usually of single-input/single-output type) between seat and control sticks motion. These are often determined with experimental campaigns, using a shaker to force the seat and measuring stick displacement [7, 8] for a number of pilots. While the fidelity of these tests may be increased by positioning the pilot in a simulator with a realistic cockpit and visual feedback, it has been demonstrated that the response is heavily influenced by several factors. The most important of them are (i) pilot physical characteristics (weight, height, and tonus) [7, 19]; (ii) cockpit configuration; and (iii) piloting task. In order to deal with these issues, some authors proposed the direct biodynamic modeling through a multibody analysis [22], while others focused their work on the influence of workload on pilot response [20]. However, due to the variability of pilot skills and characteristics and of piloting tasks, a reliable and practical approach to RPCs should include statistical considerations to model those uncertainties.

Below, several passive pilot models proposed in literature are briefly outlined. They are all presented in terms of transfer functions between seat acceleration, , and stick rotation, , namely,

Note that the number and position of poles and zeros of transfer functions change with the considered model.

3.1.1. Pilots Acting on Collective Control

Mayo’s Transfer Function. In Mayo’s experiment [7], the collective stick motion was recorded while the seat was perturbed using vertical, sinusoidal acceleration disturbances at frequencies ranging from Hz to Hz. The acceleration of pilot’s wrist was recorded using three-axis accelerometers, which were mounted on the collective grip, in a classical configuration cockpit of helicopter. The experiment was performed in an open-loop way, in that the control input provided by the pilot did not affect the acceleration of the motion platform. In order to characterize the influence of the pilot body, mesomorphic (larger size) and ectomorphic (smaller size) test pilots were considered.

The following -poles and -zeros transfer functions were identified:where is the length of the collective lever and and are two pseudointegrators (usually set to Hz in order to avoid nonphysical behavior at low frequency [11]), while depends on the reference position of collective lever, as shown in Figure 4.

Practical Biodynamic Feedthrough Models. These transfer functions, proposed in [20], provide enhanced BDFT rotorcraft-pilot models of increased accuracy and complexity equivalent to Mayo’s one. The main improvements concern the extended model frequency range and the inclusion of both the somatotype effect and the effect of task difficulty on neuromuscular admittance. Also the influence of the subject variability is included. The data have been obtained by an experimental campaign exploiting the -DOF SIMONA Research Simulator at Delft University of Technology. In the tests, acceleration and force disturbances have been applied to the control devices. The subjects were instructed to perform three disturbance rejection tasks, namely, position task, relax task, and force task.

The first task consisted in resisting the force perturbations as much as possible, while maintaining the reference position of the collective lever; for the relax task it was required to relax the arms and passively undergo the perturbations, whereas in the last task the pilot was instructed to minimize the force applied to the collective lever, yielding to the perturbation force as much as possible [20]. In these experiments, the angle of deflection of collective lever and the applied force at the inceptor were measured. In order to characterize the bandwidth of the pilot, the 0.05–21.5 Hz frequency range was considered for vertical force and acceleration disturbances, actually extending that of Mayo’s experiments (1–5 Hz).

The following three passive pilot transfer functions have been identified by the described BDFT modeling effort:

Bibby Transfer Functions. Further passive pilot models have been identified experimentally at the University of Liverpool’s Bibby flight simulation laboratory, using a visual channel flight simulator mounted upon a six-axis motion [24]. The human subject was seated in the simulator cockpit with MTx motion sensors attached near to the wrist and the elbow. Because, ideally, most of the tests required the collective lever forces to be null, the collective lever friction was set to the minimum available. A display was created with indication of the position of the control inceptors, in order to avoid drift phenomena due to the lack of feedback to the occupant. The controls (longitudinal and lateral cyclic stick, collective lever, and rudder pedals) were set to their nominal start positions and three reference positions of collective lever have been investigated (10%, 50%, and 90% of the full scale deflection). Axis excitation was a sinusoidal frequency sweep starting at Hz that steps up in Hz increments to the final value of 7 Hz. Also in this case a single-input/single-output (SISO) model has been identified, with the transfer function provided in the following zeros-poles form:characterized by a fourth-order denominator and a second-order numerator (deriving from two pairs of complex conjugated poles and one pair of complex conjugated zeros). is a gain dependent on the reference arm position. The identified values of , , , and for two pilots for three reference positions of the collective lever are reported in Table 1.

Pilot Pole #1 Pole #2 Zero Gain


Pilot Modeling for Aeroservoelastic RPC in ARISTOTEL. During ARISTOTEL European FP7 Research Project several BDFT tests have been conducted by the flight simulator at University of Liverpool [19], in order to identify pilots biodynamic response while being subject to vertical and lateral acceleration. The excitation consisted of colored noise signals, band-pass-filtered between and 10 Hz, with zero mean value and 0.004 g rms.

The motion induced to the control inceptors by the oscillations imposed to the cockpit was measured, along with the motion induced to the limbs. No specific flight task was required, the occupant being requested to hold the controls without compensating the stick vibrations and to maintain the collective lever in the proximity of the reference position to avoid excessive drift. Also in this case the pilot has been considered as a SISO system (vertical acceleration as input and wrist acceleration as output).

Pilot transfer functions have been determined in a rational polynomial form of the type

In Table 2 the transfer function coefficients identified for each pilot/test examined are presented (pseudointegrator poles, and , are set equal to 1 Hz).

1.2. ⋅ 107

Analyzing these transfer functions, it is possible to infer that there are significantly damped biodynamic poles in the range between 3 and 10 Hz. Even for the same pilot, different tests are characterized by different responses. It is worth underlining that it seems impossible to clearly distinguish professional from nonexpert pilot behavior and that there is no clear dependence on biometric measures.

3.1.2. Pilots Acting on Cyclic Control

Parham Lateral and Longitudinal Transfer Functions. An aeroelastic analysis with longitudinal/lateral pilot in the loop has been performed by Parham [8], using data from the Osprey V-22 flight test program. The initial pilot model used in that analysis was a math model, relating the pilot lateral stick displacement expressed in inches, with lateral acceleration of pilot seat, measured in . It has been refined and validated by examination of several sets of flight test data, as well as measured outputs from shake tests with pilot holding the controls. Several tests revealed a large variation of responses from different pilots and also a strongly nonlinear behavior. The following transfer functions, respectively, for lateral and longitudinal passive pilot models relating cyclic stick rotation to seat acceleration (expressed in ) have been finally determined as an optimal fit of the large amount of responses examined:where denotes the cyclic stick length.

Pilot Modeling for Aeroservoelastic RPC in ARISTOTEL. In the same experimental test campaign held at the University of Liverpool for the collective control passive pilot modeling (see Table 3), lateral control transfer functions have also been identified. Measured data revealed a good level of coherence between the lateral acceleration of the pilot seat and the lateral acceleration of the pilot wrist. Pilot’s lateral responses have been fitted using a rational transfer function composed of a 6th-order polynomial denominator and 4th-order polynomial numerator, which reads