manual flight demonstrations were performed in Ref. 1 using two .... The system was flight-tested using the Yamaha. RMAXs shown in ..... e=0 in Eq. 4 results in an command ..... Presented at the AHS 70th Annual Forum, Montreal,. Quebec ...
A System for Autonomous Rotorcraft Dual-Lift Flight Research Marc D. Takahashi, Matthew S. Whalley, and Marcos G. Berrios US Army Aviation Development Directorate Aviation & Missile Research, Development & Engineering Center Moffett Field, CA 94035
Gregory J. Schulein San Jose State University Ames Research Center Moffett Field, CA 94035
ABSTRACT A system for dual-lift research was developed and flight tested using two Yamaha RMAX helicopters equipped with customized avionics and sling-load hardware. Important features of the system were tested including precision duallift maneuvering, load tension equalization, and load motion damping. Result are presented from flight testing that include autonomous takeoff and landing, and oval racetrack pattern flying at 5 m/s carrying 171% of the rated maximum load for one aircraft. Results of formation changes while flying oval patterns are also presented as well as results on active control to equalize the sling tensions between the two aircraft. In addition, sling-load feedback results to increase the sling damping in hover are presented. Finally, the feasibility of migrating the concept to full-scale is shown in simulation using two UH-60 models in a dual-lift configuration carrying 6000 lbs at 80 kts. NOMENCLATURE δ ∆ ∆ol de/dh e χ ξ l kv, kp, ki h w η Ω τd, τv, τp Tp, Ts ωe, τe ADD DLAFCS DLSM GPS GSO NTT
Stick inputs Dual-sling configuration tilt-angle Feed-forward tilt angle Change in the sling-tension error between aircraft due to height change Sling tension error Tracking Control inputs Flight Path Command inputs Sling length Synchronized-maneuvering gains Secondary aircraft height relative to primary aircraft Dual-lift aircraft horizontal separation Base angle of dual-lift geometry Aircraft path turn-rate Synchronized-maneuvering time delays Primary and secondary sling tension Load-equalization crossover frequency and rise time Aviation Development Directorate Dual-Lift Autonomous Flight Control System Dual-Lift Synchronized Maneuvering Global Positioning System Ground Station Operator Nose-To-Tail
Presented at the American Helicopter Society 74th Annual Forum, Phoenix, AZ, May 14-17, 2018. This is a work of the U.S. Government and is not subject to copyright protection in the U.S. Distribution statement A. Approved for public release. 1
PFC QTR RTK SBS SFC SP LSF
Primary Formation Control Quartering Real-Time Kinematic Side-By-Side Secondary Formation Control Safety Pilot Load-State Feedback INTRODUCTION
The concept of moving heavy cargo with multiple helicopters has been postulated for many years (Ref. 1). Widespread use of the concept has been limited because of its high pilot workload and the acceptance that automation is necessary to make the operation manageable (Refs. 1, 2). In addition, automated duallift would require specialized sensors and hardware not commonly installed on fleet aircraft further adding to the cost of implementation. Thus, as the U.S. military’s desire to carry heavier loads has expanded over the decades, proposed solutions have tended toward the building of larger helicopters to meet the demand. Building progressively larger aircraft quickly reaches a practical size limit when considering the cost/benefit of developing specialized heavy-lift vehicles needed only occasionally for those missions where the load cannot be split up (Ref. 3). Revisiting this concept now is timely because the inexorable progression towards fully automated rotorcraft systems requires aircraft to be equipped with much of the same hardware necessary for practical multi-ship cargo operations. That is, for autonomous rotorcraft, systems such as high powered computing, accurate positioning systems, reliable
high speed ship-to-ship communication, and range sensing can be dual purposed for the dual-lift task further reducing the cost of deployment. The dual lift problem has been studied extensively over the years beginning with the development of the equations of motion, such as in Ref. 4. Stability and control of dual lift was analyzed using methods ranging from classical feedback architectures to more sophisticated nonlinear feedback techniques (Refs. 510). The stability of dual-load dynamics was explored in a wind tunnel test in Ref. 11. Optimized formation control was studied in Ref.12 with a simulation of four UH-60s carrying a load. Limited manual flight demonstrations were performed in Ref. 1 using two CH-54B Tarhe Flying-Crane helicopters. More recently, multi-lift using three 12.5 kg helicopters performing linear vertical and horizontal translation maneuvers with a 5 kg load was demonstrated in hover (Ref. 13). This paper describes a system to advance the previous described multi-lift research by flight demonstrating autonomous dual-lift operations on a reduced scale using the two autonomous RMAX helicopters shown in Figure 1. These 165 lb helicopters were successfully used as a development research platform (Ref. 14) at the Aviation Development Directorate (ADD) prior to transferring the resulting autonomy technology to the Rotorcraft Aircrew Systems Concept Airborne Laboratory (RASCAL) JUH-60A Black Hawk (Refs. 15 and 16). Therefore, they are a proven surrogate testbed with which to demonstrate a dual-lift concept prior to fullscale implementation.
The ultimate goal of this research is to develop and flight validate a dual-lift control system that can be confidently scaled to full-scale helicopters and provide three basic features: synchronized maneuvering in tight formations, equalization of sling-load tensions, and increased sling-load damping. The scope of this work is limited based on the importance of these feature’s role in realizing a practical dual-lift system. The feature related to maneuvering is essential to the dual-lift task and is built into the approach presented here. Load equalization is also important as well, but only a basic example of how this concept can be executed in the control system will be demonstrated. Finally, the sling-load damping feature, though not essential, is of interest as part of managing the load, and some configuration specific examples will be shown. Therefore, the dual-lift control presented here is the basic system to achieve the goals of the work, but also provides enough flexibility to allow the loadequalization and load-damping features to be replaced will alternate implementations. So, the presented system not only acts as a surrogate for fullscale development, but also was designed to serve as a platform to explore alternate methods of load control, which was done in Ref 17 in developing general load-equalization and load-damping methods. The paper begins with a description of the Dual-Lift Autonomous Flight Control System (DLAFCS). Next, the system hardware is described followed by a discussion of how the dual-lift system was tested. Flight test results are presented for the formation maneuvering, load equalization, and load damping control features of the control law. Finally, simulation results on application of the approach to full-scale helicopters are presented followed by conclusions. DUAL-LIFT AUTONOMOUS FLIGHT CONTROL SYSTEM (DLAFCS)
Figure 1: Aviation Development Directorate (ADD) Yamaha RMAXs.
2
The Dual-Lift Autonomous Flight Control System (DLAFCS) is shown in Figure 2 and is implemented by modifying the single-ship AFCS described in Refs. 14-16. The primary aircraft is represented at the top of the figure and is referred to as the “primary”. The secondary aircraft is represented at the bottom of the figure and, likewise, will be referred to as the “secondary”. The overall system concept assumes there is a designated primary commanding a secondary in a synchronized manner. The result is a system that can perform coordinated maneuvering needed for practical dual-lift operations.
Figure 2: Dual Autonomous Flight Control System. Single ship Autonomous Flight Control System Before describing the dual-lift system, the single ship Autonomous Flight Control System (AFCS) is described using the primary system at the top of the Figure 2. There are two guidance methods to direct the AFCS shown in the upper left corner of the figure: Path Generation and Vector Command. The Path Generation method allows the operator to specify a predetermined path by entering a few widely separated waypoints along with macro-level parameters such as maximum speed, climb rate, and acceleration limits. These waypoints are spline-fitted into a smoothed set of trajectory data that is sent to the waypoint-trajectory interface of the Waypoint Control block. The Vector Command method receives discrete changes to the desired commanded velocity in the form of speed, glide slope and heading rate change and generates a stream of vector commands that are sent to the vector-command interface of the Waypoint Control block.
3
The Waypoint Control has a waypoint-trajectory interface that accepts a series of smoothed waypoints from the Path Generation that includes position, velocity, acceleration, and time information to navigate the path. It also has a vector-command interface that accepts a time sequence of velocity commands from the Vector Command process. In both cases, the Waypoint Control converts the guidance information into a stream of inertiallyreferenced trajectory data (χ) sent to the Tracking Control. The Tracking Control tracks the inertial reference path signal (χ) generated from the Waypoint Control and passes aircraft-referenced command signals (ξ) to the Inner-Loop Control. The Tracking Control acts as the autopilot around the Inner-Loop Control and has a hover mode and forward-flight mode, which transitions between them at a fixed speed threshold. The experimental work with the RMAXs described herein is confined to hover low speed, so the forward flight mode is not used. The simulation results presented later using the UH-60 models, however, does use the forward flight mode.
The Inner-Loop Control provides the primary attitude and flight path stabilization of the system and uses the aircraft-referenced outputs (ξ) from the Tracking Control as its inputs and generates the four de-mixed lateral, longitudinal, pedal, and collective stick positions (δ). For the RMAX, the Inner-Loop Control also regulates the engine RPM through the throttle.
load equalization control. Unlike the primary, the secondary is always in Vector Command mode and is controlled by the primary. As with the PFC, the SFC has a default configuration that emits constant bias inputs relative to the primary aircraft, but alternate behaviors can be inserted into the block as will be shown later in “Side-by-Side Load Equalization”.
Modifying the AFCS for Dual Lift
The primary and secondary Inner-loop Control blocks in Figure 2 were modified with the Load State-Feedback (LSF) blocks, which share information about the other vehicle’s load state. Therefore, LSF control laws can be implemented to damp the swinging motion of the load using the load state of the entire system to command the respective lateral and longitudinal stick commands.
The objective of the system development was to demonstrate three key features of dual-lift: maneuvering in dual-lift formations, load equalization control, and sling-load damping control. To support the first two, the AFCS was modified assuming a primary vehicle that is commanded as the single aircraft version of the AFCS. That is, the primary vehicle can be directed using either the Vector Command or Path Generation method. The primary commands a secondary vehicle by sending vector commands to the vector-command interface of the secondary Waypoint Control. The load damping control feature gives the primary and secondary access to each other’s load state, so actuator commands issued by the respective aircraft are based on the entire system’s load-state. The primary Waypoint-Control was modified in Figure 2 with a Primary Formation Control (PFC) block to control the position of the secondary with respect to the primary aircraft. The default mode for this block is a simple bias control system, but the PFC can be used to accommodate alternate control implementations. The PFC commands large relative position to arrange the configuration of the formation. For example, placing the secondary at an angular position behind the primary at a specified distance. The PFC is aware of the position of the secondary and the load state, and is intended to command in an open-loop feed-forward way to adjust the formation of configuration. The secondary Waypoint Control was modified with the Secondary Formation Control (SFC) block (Figure 2) that allows the secondary to move itself with respect to the position commanded by the primary. The commands from the SFC are meant to be small orthogonal position changes relative to the PFC configuration commands. Used in this way, the SFC acts as fine position control (feedback), while the PFC commands larger, macro, configuration controls (feed forward). Since SFC is aware of the position of the primary and load state it can make adjustments to its position to control such measures as the difference between the sling tensions; i.e. for 4
Dual-Lift Synchronized Maneuvering (DLSM) The implementation of the Dual-Lift Synchronized Maneuvering (DLSM) is shown Figure 3. As mentioned previously, the overall approach is to command the primary via waypoint or velocity command and have the primary command the secondary via the vector command interface. The secondary is commanded by a velocity vector, but it also needs the desired command position to avoid the two aircraft positions from drifting apart. This will happen if the secondary only received velocity commands and relies on integration to hold a relative position. To prevent this drift, the flight path position of the secondary is synchronized in the Waypoint Control block to the position sent from the primary. Commanding both aircraft in a dual-lift formation starts at the primary where vector commands or waypoint commands enter the Waypoint Control block shown in the top of Figure 3a. These commands are transformed and smoothed into inertial coordinates and are delayed by τd seconds and sent to the primary tracker (χ). The inertial commands are also sent to the secondary with adjustable delays as well after commands from the PFC block are added. The primary delays the command to the tracker to allow for command delays to propagate to the secondary. The other delays τp and τv allow the secondary command positions and velocities to be adjusted relative to the primary delay. Delays of approximately τd=τv=2 sec and τp=0 worked well since the position command response of the tracking is delayed by about 2 seconds. In other words, this says to hold back the primary commands and secondary position, while immediately sending secondary velocity.
Figure 3: Synchronizing the Primary and Secondary Waypoint Control. The commanding from the primary represents the feed-forward part of the maneuver synchronization, but the feedback part is in the secondary Waypoint Control block at the bottom of Figure 3a. The secondary buffers the commands in the synchronization block that ensures the times stamps of the commands are consistent between the two aircraft. This is done to avoid large position transients, which can be caused by any large inadvertent time delays in the communication between the aircraft when they are in motion. The secondary is configured always in vector command mode and accepts the velocity commands 5
from the primary as it normally would in single aircraft AFCS mode. However, the primary internal position commands are also relayed to the secondary and the error between the integrated velocity position and commanded position is reduced with a limited proportional-integral feedback shown in Figure 3b. Gain values of kv=4.8 1/s, kp=3.0 1/s2, ki=0.4 1/s3, and velocity limit of 2 m/s and acceleration limit of 2 m/s2 were used for the flight test. The limited PI block also has a feature to extrapolate the position to avoid stair-stepping position commands from entering the tracker. This is necessary because the control law runs at a much higher rate than the command data from the primary. The control law
runs at 50Hz but the velocity commands from the primary are only sent at 5Hz to keep the data transfer bandwidth low. The velocities and positions are then sent to the secondary tracker after they are summed with SFC commands.
only communicates with the ground station. To ensure time synchronization with the other aircraft, the computer clocks on both aircraft were kept in synchronization with GPS time.
As configured, the DLSM has an inherent delay of about two seconds due to the τd=2 sec, but this is not a significant penalty since dual-lift operations are expected to be fully autonomous and the time scale of a load transport operation is considerably longer than this delay. In addition, the approach should work with multiple aircraft given one primary and several secondaries for multi-lift or multi-aircraft formation control. This delay also provides future states that can be leveraged by alternate control methods such as Model Predictive Control (MPC). Reference 17 used this future state to calculate optimal secondary trajectories using nonlinear programming (NLP) methods to make the PFC block act as a feed-forward to a load equalization controller.
TEST EQUIPMENT Aircraft The system was flight-tested using the Yamaha RMAXs shown in Figure 4a, which were modified by the Aviation Development Directorate (ADD) with custom built avionics payloads. This commercial platform was originally developed for remotely piloted agricultural seeding and spraying and has a stabilizer bar on a two bladed teetering rotor with an approximate 10 ft diameter. The modified aircraft weighs165 lbs with a full, 1.6 gallon, tank of gas-oil fuel mix and is capable of approximately 50 min of hover flight using its 21 HP two-stroke engine. Given this, under the conditions of the flight testing (0 MSL, 22.2 degC) the aircraft has approximately 35 lbs of useable payload (Ref. 18). The avionics payload bay (Figure 4b) houses a NovAtel Span OEMV/LN-200 unit that provides aircraft position, velocity, accelerations, attitude, and attitude rate measurements. The GPS in this unit receives correction from the ground station and can achieve a centimeter-accurate differential Real Time Kinematic (RTK) solution. A 64-bit Core 2 Duo PC/104 computer cycles at 1.2GHz (VersaLogic VLEPM-35E, extended temperature) and runs the DLAFCS software on the CENTOS-7 Linux operating system. The bay houses an IP radio, which is used for telemetry/network communication and 6
Figure 4: Autonomous Dual-Lift Hardware.
Figure 4c shows the bottom side of the avionics payload which is attached to the vehicle underside. In addition to the payload bay, there is an ATI Industrial Automation FT15812 six-axis force/moment transducer with an interface control box, which is used to acquire measurements to estimate the sling tension and angles. Inline force/moment measurement limits are a maximum z force = 60 lbf, x-y force = 30 lbf, and moments = 40 lbf-in at 7000 Hz. The unit data rate was configured with an antialias filter at 5 Hz and data decimated to 50 Hz.
operational system would use a devoted high reliable, low latency link. For this test, each aircraft was exchanging data at a 5 Hz update rate, which was approximately 12 Kbps of data.
A custom electromagnet was attached to the force transducer, which, when energized, holds a metal cap attached to the sling. The electromagnet is centered 0.3 m under rotor and can hold approximately 120 lbs of downward force. The electromagnet is energized and de-energized to attach/release the sling and is controlled via software, safety pilot transmitter, or ground crew payload switch. Figure 4d shows the metal cap used to attach the load to the electromagnet. A stainless steel captive pin bow shackle (rated 3310 lb) attaches to the cap and 15mm tubular nylon webbing sling (rated 2300 lb), which was cut to give a sling length of 40 ft. Also shown are the 20, and 40 lb load sand bags, which have 10 inch diameter and 12-20 inch height. The sand bags allowed the weight to be easily adjusted and can handle being dropped short distances without significant damage. Ground Station The aircraft are controlled from the ground station shown in Figure 5, which is equipped with a computer running the CENTOS-7 Linux operating system and three terminals. Ground station software controls a NovAtel GPS base station, tracking cameras, display, and weather station software. Dual tracking cameras are mounted on pan-tilt units along with directional antennas on the top of the station. Communication with the aircraft is through 902-928 MHz IP radio modems and is configured to provide 230Kbps data rate, and has an approximate 50-100 msec ping time from aircraft to ground. In addition to state data between the aircraft, GPS corrections are sent to both aircraft over this link. The ground station clock was also set based on the GPS time to ensure time synchronization with the aircraft. For this test, data exchanged between the aircraft went through the ground station, which would always be near both aircraft and was assumed to provide enough bandwidth for all shared aircraft data. An 7
Figure 5: Ground station.
OPERATIONS Test Area The system was tested at the NASA Federal Airfield at Moffett Field, California using the north end of the east runway and taxiway (Figure 6). This area provided a large enough maneuvering space with a hard surface to support a chase vehicle to allow monitoring of the aircraft in flight by safety personnel.
stayed on the paved portion of the airfield. An airfield ground safety observer was always in contact with tower during operations and the SPs were always in radio contact with the ground station. The aircraft maneuvers were always arranged so the secondary aircraft was abreast or behind the primary aircraft with the primary closest to the chase vehicle. Pattern flying was restricted to the oval pattern show in the figure, which always kept the primary on the outside of the pattern and the secondary on inside and never ahead of the primary. These restrictions always gave the pilots a sense of aircraft separation and avoided confusion caused by aircraft switching positions or one aircraft blocking the view of the other.
Addition Dual-Lift Risk-Reduction Measures To reduce risk, simulation was used to rehearse the test cards prior to flying. Both a desktop simulation of the aircraft was used as well as a hardware version that ran the software on the aircraft flight computers attached to the aircraft and could engage and disengage the system using the radio transmitters. The hardware and desktop simulation capability described in Ref. 14 was modified to accommodate dual-lift dynamics assuming a point mass load.
Figure 6: North end of Moffett Airfield east runway and taxiway.
Safety Pilots Risk mitigation required safety pilots (SP) for each aircraft, who could disengage the control system and manually fly their respective aircraft using handheld transmitters. In hover, the pilots generally preferred standing behind the aircraft. When in the low-speed flight, the pilots were seated in an open-air electric chase vehicle in the right front (primary SP) and right rear (secondary SP) seats facing out the side of the vehicle. This arrangement allowed easy verbal communication between pilots and the driver. The driver followed the moving aircraft pair slightly behind and to the left of the primary and kept an acceptable view for both pilots and simultaneously 8
Additional software features were added to help reduce the chance of aircraft collision. At the output of the PFC and the SFC in Figure 3a, limiting was implemented to prevent inadvertently commanding the aircraft too close to each other. When in motion, SPs or ground station operators (GSO) could stop the commanded motion of the pair, without disengaging the control laws. This allowed GSO operators or SPs to stop the motion at any time if anyone suspected anomalous behavior that was not perceived as an imminent danger. This was an important feature as the cost of disengagement was high, since reestablishing the configuration to an engaged state took several minutes. An automatic load-drop feature was added so any disengagement or load drop on one side would cause the load on the other aircraft to be released. This feature was only available as long as the communication link was present, but was important to avoid having only one aircraft holding the load by itself.
Configurations After considering the safety issues mentioned above, a set of configurations were defined to test the duallift maneuvering and are shown in Figure 7. The configurations always kept the secondary behind or on the inside of the primary during the oval pattern maneuver, which allowed for the best perspective for each SP. The Side-By-Side (SBS) configuration (Figure 7a) arrange the aircraft to move abreast, where the line between them is 90 deg from the flight path vector. The Quartering (QTR) configuration (Figure 7b) has the secondary 45 deg behind the starboard side of the primary. Finally, the Nose-ToTail (NTT) configuration (Figure 7c) has the secondary directly behind the primary. Engaging the System Engaging the aircraft in-flight always has the potential for transients due to the experimental nature of the system. In addition, there are high risks associated with the load touching the ground while the aircraft is moving or when near another aircraft. Therefore, engagement in the air with the sling and load attached was considered problematic, since the 40 ft slings did not have sufficient slack when the aircraft were at a safe engage altitude above 32 ft AGL. Furthermore, the sling attached to both aircraft caused a situation where they needed to be a relatively close to each other during engagement. Based on this assessment, it was decided there was less risk with engaging the aircraft on the ground and taking off with the loads, where the control loops are open and are closed after lift-off. In addition, a fullscale system would need to take off and land autonomously with the load, so there was value in demonstrating this capability as it would be done in a full-scale operational system. To engage the system, the aircraft were positioned in a SBS configuration on the ground with the load attached ahead and centered in front of the aircraft. The SBS gave best view for the SPs who were standing behind their respective aircraft. The GSOs then configured the DLAFCS software for engagement on the ground and instructed the primary SP to engage followed by the secondary SP. Figure 7: Dual-lift configurations.
9
Figure 8: Autonomous dual-lift takeoff.
10
Figure 9: Autonomous dual-lift landing.
RESULTS Auto Takeoff and Landing An autonomous takeoff is shown in Figure 8 and began with both aircraft in a SBS configuration with the engines idling. The DLAFCS was engaged with control loops opened, and the load was centered and ahead of the aircraft. (Figure 8a). During takeoff a significant risk was entanglement with the attached slings as the rotor downwash was high enough to move them. Light weights were place on the sling near the aircraft to prevent them blowing into rotors or over the skids. During the takeoff, the geometry was also arranged to prevent pulling the sling too tight, thus leaving enough slack to handle emergency disengage situations or transient motions. The primary proceeded with the rotor spin-up and lift-off sequence to 5 ft AGL (Figure 8b) with control loops closed. After lift-off, the aircraft was moved to a safer 10 ft AGL (Figure 8c). The secondary then spun-up and then lifted off (Figure 8d), after which it moved to the same AGL as the primary (Figure 8e). The DLSM was enabled and the pair then moved forward over the load together (Figure 8f), and ascended vertically to lift the load off the ground (Figure 8g). The autonomous landing began with the loaded aircraft hovering in the SBS configuration (Figure 9a) with the safety pilots behind. The load was then lowered to the ground and the slings were released from both aircraft (Figure 9b). The DLSM was disabled and the primary then backed over the landing point (Figure 9c) and proceeded to descend to land and idled its engine (Figure 9d). The secondary then backed over its landing point (Figure 9e), descended to land, and wound down its rotor (Figure 9f). Dual-Lift Maneuvering To reposition the system, the GSO commanded the primary. As a general rule, during hover, the aircraft were pivoted in place about the primary to the desired heading and then they were commanded forward to the destination. Turn maneuvers were restricted to a maximum turn rate of 9 deg/sec for the aircraft in a SBS configuration at 40 ft separation. Basic testing of the SFC was done by inserting step inputs to change the relative position from the primary commanded position. The PFC was initially tested in hover by commanding the secondary from the PFC block to the QTR, NTT, and SBS configurations.
11
Once checked out in hover, both aircraft flew the clockwise oval patterns shown in Figure 10 at 5 m/s. The pattern flights started and ended in the SBS configuration at 40 ft separation (4 rotor diameters) with 40 ft slings. The figure shows three configuration, SBS, QTR and NTT, where the latter two convert from the SBS during the maneuver so the aircraft flew in the changed configuration during the sustained 180 deg turn at the south end of the field. Changing the formation in flight demonstrated the versatility and precision of the proposed method. Figure 10a shows results for the SBS configuration, which shows the primary (blue circle) body-fixed nose-forward (green) and perpendicular to the right of the aircraft (blue) vectors. The secondary (red circle) is held in a position perpendicular to the flight path (magenta dotted line). The symbols are depicted at various time points during the maneuver and it can be seen the aircraft kept a tight SBS formation during the oval pattern. Ping times from the aircraft to the primary and secondary were continuously monitored by the GSOs and, generally remained below 100 msec. The time delay of inter-aircraft communication is critical to maintaining formation so it was monitored by ground personnel during the oval pattern flying. For the QTR configuration in Figure 10b, the pair started in the SBS configurations, and, after turning the first corner, the PFC commanded the secondary aircraft into the QTR position. The pair made the sustained 180 deg turn at the south end of the pattern while in the QTR configuration. After exiting the turn, the PFC changed the formation back to the SBS configuration for the final turn and ended in that formation. Tight formation was retained during the maneuver. For the NTT configuration in Figure 10c, after the pair turned the first corner, the PFC commanded the secondary aircraft into the NTT position, which was the formation held as the pair made the sustained 180 deg turn at the south end of the pattern. After exiting the turn, the PFC changed the formation back to the SBS configuration for the final turn. As with the previous cases, tight formation control was retained during the maneuver. The NTT formation has the highest risk of collision, since any sudden slowing of the leading primary could cause a rear-end collision from the secondary. The SBS and QTR formations have enough lateral offset to reduce this collision risk.
Table 1 shows the relative error of the SBS position when the oval pattern was flown with different load weights. The RMS error for each axis remained about the same for each run. The maximum vertical error was less than 13 cm while the horizontal sided error stayed below 26 cm and the fore-aft error stayed below 39 cm. The error is slightly higher for the unloaded case, which is probably due to the aircraft being more susceptible to turbulence when lightly loaded. The error results demonstrates the effectiveness of the DLSM approach to keeping the aircraft in a tight formation during maneuvers. Table 1: Position error during Side-By-Side oval maneuvering with different load weights. Run
Load Wind Fwd Err Side Err Vert Err (lbs) m/s (m RMS) (m RMS) (m RMS)
17022.009
0
2.57
0.385
0.259
0.091
17022.012
20
1.78
0.378
0.218
0.120
17029.018
40
2.91
0.328
0.191
0.124
171006.027 60
2.46
0.326
0.171
0.113
Side-by-Side Load Equalization The previously described maneuvers are open-loop commanding of the aircraft, but the PFC and SFC components can also be used as an active control to equalize the tensions of the slings. To demonstrate equalization, a simple, closed-form pre-filter and feedback solution, based on the geometry, flight speed, sling arrangement, and point load mass was implemented and tested. The equalization control law is based on Figure 11a, which shows a load suspended at two points in a SBS configuration in a steady turn. The objective is to equalize the load so the error between the two tensions is zero. 𝑒𝑒 = 𝑇𝑇𝑝𝑝 − 𝑇𝑇𝑠𝑠
(1)
The error is controlled by adjusting the angle of the entire sling configuration, ∆, which is approximated by changing the height of the secondary relative to the primary divided by the width of the sling configuration. Figure 10: Dual-Lift configurations and conversions from Side-by-Side to Quartering and Nose-to-Tail formation while flying the oval pattern at 5 m/s.
12
𝛥𝛥 =
ℎ
𝑤𝑤
(2)
In a steady turning flight, the sideward acceleration of the load is given by
𝑎𝑎 = 𝑉𝑉 𝛺𝛺 + 𝛺𝛺2 𝑙𝑙 𝑐𝑐𝑐𝑐𝑐𝑐 𝜂𝜂 + 𝛺𝛺2 𝑙𝑙𝑙𝑙 𝑠𝑠𝑠𝑠𝑠𝑠 𝜂𝜂
(3)
so, the error can be approximated by 𝑒𝑒 =
𝑚𝑚 𝑡𝑡𝑡𝑡𝑡𝑡 𝜂𝜂 𝑠𝑠𝑠𝑠𝑠𝑠 𝜂𝜂
�𝑉𝑉 𝛺𝛺 + 𝛺𝛺2 𝑙𝑙 𝑐𝑐𝑐𝑐𝑐𝑐 𝜂𝜂 + 𝛥𝛥 (𝛺𝛺2 𝑙𝑙 𝑠𝑠𝑠𝑠𝑠𝑠 𝜂𝜂 − 𝑔𝑔)�(1 − 𝛥𝛥)
turn. The feed forward helps by preventing the feedback from having to chase the error during transients of the configuration. The bottom plot shows the difference between the sling tensions both with and without the equalization.
(4)
Taking the derivative gives a first order approximation of a model of how the error changes with respect to changes in height. 𝑑𝑑𝑑𝑑
𝑑𝑑ℎ
=−
𝑚𝑚 𝑡𝑡𝑡𝑡𝑡𝑡 𝜂𝜂 𝑤𝑤 𝑠𝑠𝑠𝑠𝑠𝑠 𝜂𝜂
(𝑔𝑔 − 𝛺𝛺2 𝑙𝑙 𝑠𝑠𝑠𝑠𝑠𝑠 𝜂𝜂 + 𝛺𝛺𝛺𝛺+𝛺𝛺2 𝑙𝑙 𝑐𝑐𝑐𝑐𝑐𝑐 𝜂𝜂) (5)
Which is used to construct the PI control law shown in Figure 11b.
A feedforward is implemented based on commanding the height change to balance the load. That is, setting e=0 in Eq. 4 results in an command 𝛥𝛥𝑜𝑜𝑜𝑜 =
ℎ𝑜𝑜𝑜𝑜 𝑤𝑤
=
𝑉𝑉 𝛺𝛺+ 𝛺𝛺 2 𝑙𝑙 𝑐𝑐𝑐𝑐𝑐𝑐 𝜂𝜂 𝑔𝑔−𝛺𝛺 2 𝑙𝑙 𝑠𝑠𝑠𝑠𝑠𝑠 𝜂𝜂
(6)
which is the open loop estimate of how to change the height to zero the error. The feedback of the control law changes the height of the secondary to zero the load error and has a proportional and integral gain feedback represented in terms of a crossover frequency, ωe, and a lead compensation with time constant τe. For this work, the cross over frequency was set to ωe= 0.1 rad/sec and the time constant was set to τe=0.0 sec. The time constant can be used to add addition lead in the feedback, however, this was unnecessary, since a relative low crossover was needed to achieve the equalization. A more in-depth analysis of the dual load dynamics and how to design the load equalization is provided in Ref. 17. The pre-filter was implemented in the PFC block to command the secondary aircraft height, and the feedback in height was implemented in the SFC block. Figure 12 shows the sling tensions for the primary and secondary aircraft during a SBS oval pattern with and without load equalization carrying a 40 lb load. The top time history shows how the tensions change relative to each other as the vehicles maneuver without tension equalization, where the difference reach approximately 5 lbs. In the middle plot, the effect of the PFC/SFC load equalization is shown to be very effective in keeping the tension the same. Because of the low cross over frequency, the feedforward is necessary to help the feedback keep the loads equalized as the aircraft enters and exits the 13
Figure 11: Side-by-Side load equalization control.
Dual-Load State-Sling Feedback Damping out the swing motion of a sling load is a useful feature to decrease the time to place a load (Ref.19). The LSF blocks in Figure 2 provide the means to achieve this and were tested by using a simple position and rate load state feedback shown in Figure 13. For simplicity, and to check both axes of control, the LSF was tested for the SBS and NTT configurations separately. Each aircraft has the other aircraft’s load state, so the feedback is applied based on measurements from the entire sling load configuration.
Figure 13: Dual-Lift load state feedback for the SBS and NTT configurations.
Figure 12: Sling tensions with and without load equalization carrying a 40 lb load in a SBS configuration flying the oval pattern at 5 m/s. The feedback is based on the single load topology from Ref. 20, but uses averaged sling angle and sling angle rate from both aircraft and was expected to increase the load damping in the respective axis. For the NTT configuration, the lateral angle is used to regulate the side to side swing motion. For the SBS, the longitudinal angle is used to regulate the fore and aft swing angle. Load state feedback gains for a 30 ft sling from a single load analysis was used to test the implementations of the LSF blocks. For the longitudinal control, the gains were kp=0.1764, kd=0.1754 sec, Tlon=1/3 sec, and for the lateral control they were set to kp=0.7244, kd=0.3368 sec, Tlat=1/3 sec =, where the stick commands are nondimensionalized to +/-1. Although not optimal for this configuration, these gain were sufficient to validate the operation of the system. An analysis of how to select and analyze dual-lift sling angle feedback for the general configuration is provided in Ref. 17.
14
To measure the effect of the feedback, a chirp input was injected into the respective control axis of the secondary aircraft to excite the load swing motion. The closed-loop frequency responses were calculated using the CIFER© frequency response identification methods from Ref. 21. Figure 14a shows the lateral open and closed load damping for the system in the NTT configuration. The resulting frequency analysis shows an increased level of load damping as the peak magnitude drops 6dB with the load state feedback for both the simulation and flight cases. Correlation with the simulation below 2 rad/sec results is reasonable. Figure 14b shows the longitudinal open and closed load damping for the system in a SBS configuration. Damping was higher in the longitudinal axis, so a much smaller increase in damping is seen. This effect is similar to the single sling case, which has much more damping in the longitudinal axis than in the lateral axis. Correlation with the simulation results is also reasonable. The results show that the LSF concept as implemented can be used to increase the damping of the load. The lateral swing activity is higher in the NTT configuration and the damping can be increased with LSF. However, the inherent longitudinal damping is higher in the SBS configuration, so only a modest amount of additional damping can be gained with the LSF.
Figure 14: Frequency response in the sling angle on the secondary aircraft in the a) lateral and b) longitudinal direction. Full-Scale Dual Lift Simulation To demonstrate the feasibility of using the DLAFCS in full scale, a simulation of UH-60 Black Hawk helicopters flying a SBS configuration was performed. The simulation model is described in Ref. 15 and consists of linearized UH-60 models at 0, 40, 80, and 120 kts. The stability derivatives are interpolated across the 0 to 120 kt speed regime and are combined with full non-linear kinematics. Bounding trim points are set to preserve stability derivatives that depend on forward velocity. The UH60 model was also used extensively to validate the implementation of the full scale autonomy in Refs. 15 and 16. The UH-60 simulation shares a significant portion of its code base with the RMAX simulation including the point-mass dual-lift model and the DLAFCS architecture. Figure 15a shows the two aircraft with a 120 ft (~2 rotor diameters) separation and a 180 ft sling carrying a 6000 lbs in a SBS configuration. The AFCS gains are the same as those used in the flight test described in Ref. 16. The turn coordination was configured to use inertial velocity to zero out the lateral velocity. Accelerometer-based turn coordination for dual lift requires further analysis, but would likely involve 15
using available force measurements. The limited PI controller gains to synchronize the aircraft positions are the same for the RMAX, but the limits were gain scheduled as shown in Table 2. Table 2: Limited PI control for position synchronization for UH-60. kp ki Speed kv Vel Lmt Acc Lmt (kts) (1/s) (1/s2) (1/s3) (m/s) (m/s2) 0
4.8
3.0
0.40
6.0
2.0
40
4.8
3.0
0.40
9.0
4.0
80
4.8
3.0
0.40
12.0
6.0
120
4.8
3.0
0.40
15.0
8.0
Figure 15b shows the path of both aircraft flying a pattern where the aircraft starts in hover and accelerates to 80 kts. The pair turn at 2 deg/sec until the heading changes 180 deg followed straight and level flight for 30 sec. The pair turn at 3 deg/sec changing heading another 180 deg followed by a leveling and a deceleration to hover. The commanded heading rate change was limited to 1 deg/sec/sec to avoid large transients in the forces as the pair entered and exited the turns.
Figure 15: Simulation of Dual UH-60 helicopters carrying 6000 lbs at 80 kts.
Figure 16 shows the sling tensions for two cases, the top plot without load equalization, and the middle plot with equalization. The bottom chart shows the difference between the primary and secondary tensions. Without the load equalization, the loads become substantially different reaching a difference of 4000 lbs. With the load equalization, the load differences are kept to within 250 lbs of each other. The RMS error of the secondary aircraft position were radial=2.5 m, forward=1.2 m, height=0.5 m. REMAINING ISSUES RELATED TO ACHIEVING FULL-SCALE DUAL-LIFT The flight demonstration and analysis of the dual-lift concept described herein did not address some important, though not insurmountable issues. The GPS solution used the high precision, centimeteraccurate RTK solution to position the aircraft. On a full-scale system, a GPS system using a moving base station, such as the technique described in Ref. 22, could be used to get an accurate relative position of the primary to the secondary. This system may not necessarily have accurate absolute position of the primary, but the relative position will be accurate 16
enough to avoid aircraft collisions. Using GPS also provides a simple means of clock synchronization as was done in this work. Not using any GPS would require using some sort of local measurement system between the aircraft, such as sonars or laser range finders. The relative position of the secondary to the primary could then be used with the primary’s position to accurately determine the secondary’s relative position. For simplicity, communication between the aircraft was done through the base station where the downlink delays were kept below 100 msec. On an operational system, direct aircraft-to-aircraft communication would be required and the delay would be have to be kept to a minimum at the higher flight speeds. For this system, approximately 12 Kbps of data per aircraft were being sent to coordinate the dual-lift system. This included the load state measurements as well so the bandwidth requirements could be significantly reduced if LSF is not required.
1.
2.
3.
4.
Figure 16: Sling tensions of Simulated Dual UH60 helicopters carrying 6000 lbs at 80 kts. Finally, the full-scale simulation assumed inertial velocity base turn coordination, which would not be as effective in an operational system. Therefore, either an air data system would be required to turncoordinate or the acceleration-based turncoordination would need to be modified to handle the side loads from the slings. Since a load measurement system is required to equalize the load tensions, these forces would be available to bias the side-force accelerometers to turn coordinate. CONCLUSIONS A system for dual-lift research was developed and flight tested using two Yamaha RMAX helicopters. The system demonstrated three key features of duallift research: maneuvering in tight formations, slingload tension equalization control, and sling load damping control. Elements of realistic operational maneuvers were demonstrated such as autonomous takeoff and landing, oval racetrack patterns, and formation changes while in low speed flight. Load equalization was demonstrated in forward flight as well as load state feedback to increase the dual slingload damping in hover. Finally, dual-lift simulation with UH-60 aircraft maneuvering in hover and 80 kts showed the feasibility of full-scale dual-lift. The conclusions from this work are as follows. 17
5.
The Dual-Lift Autonomous Flight Control System (DLAFCS) was effective in managing dual-lift operations with the RMAX helicopters performing autonomous takeoff and landing, and realistic maneuvering carrying 171% of the rated load for one RMAX aircraft. The Dual-Lift Synchronized Maneuvering (DLSM) approach was effective in coordinating the flight of the RMAX helicopters. Precise coordinated maneuver control between the aircraft was shown while flying oval patterns with position errors below 26 cm laterally, 39 cm forward, and 13 cm vertically. A simple side-by-side load equalization algorithm to balance the load between aircraft was successfully demonstrated while flying oval patterns at 5 m/s. Increased damping for the dual-lift SBS and NTT configurations was demonstrated using load state feedback. More improvement in damping appeared achievable for the lateral axis in the NTT configuration than in the longitudinal axis in the SBS configuration. However, the SBS had more inherent longitudinal damping. The feasibility of dual-lift at full-scale was demonstrated by simulating UH-60 helicopters maneuvering from hover to 80 kts carrying a 6000 lb load in a SBS configuration. The same load equalization control law as was used on the RMAX helicopters and was show to be effective in keeping the simulated UH-60 load tension differences below 250 lbs. ACKNOWLEDGEMENTS
The authors would like to acknowledge the support of the safety pilots Mr. Perry Kavros, and Mr. Jesse Kavros whose piloting skills and insights were critical to the success of the test. We also would like to thank Mr. Phillip Schuyler who was critical in building the custom hardware used on the RMAXs, maintaining the aircraft flying reliably, and supporting the test operations. REFERENCES 1. Meek Jr., T. and Chesley, G., “Twin Helicopter Lift System Study and Feasibility Demonstration,” Sikorsky Engineering Report 64323, prepared for US Army Aviation Material Directorate, December 1970.
2. Korsak, K., Meenen, K., Meyers, D., and Piasecki, F., “Multi-Helicopter Heavy Lift System Feasibility Study,” NAVAIR Contract No. N62269-71-C-0581, Report No. 39-X-11, February 1972. 3. Meier, W. and Olson, J., “Efficient Sizing of a Cargo Rotorcraft,” AIAA Journal of Aircraft, Vol. 25, (6), 1988. 4. Cicolani, L. and Kanning, G., “Equations of Motion of Slung-Load Systems, Including Multi-lift Systems,” NASA TP 3280, 1992. 5. Curtiss, H. C. and Warburton, F. W., “Stability and Control of the Twin Lift,” Journal of the American Helicopter Society, Vol. 30, (2), April 1985. 6. Hess, R. and Tran, P., “Pilot/Vehicle Analysis of a Twin-Lift Helicopter Configuration in Hover,” AIAA Journal of Guidance, Navigation, and Control, Vol. 11, (5), September–October 1988. 7. Prasad, J., Schrage, D., and Mittal, M., “Comparison of Nonlinear Controllers for Twin-Lift Configurations,” AIAA Guidance, Navigation, and Control Conference, August 1989. 8. Menon, P. and Prasad, J., “Nonlinear Control of a Twin-Lift Helicopter Configuration,” Journal of Guidance, Control, and Dynamics, Vol. 14, (6), November–December 1991. 9. Song, Y., Horn, J., Li, Q., and Langealaan, “Modeling, Simulation, and Non-Linear Control of a Rotorcraft Multi-Lift System,” 69th Annual Forum of the American Helicopter Society, May 2013. 10. Berrios, M.G., Tischler, M.B., Cicoloni, L.S., and Powell, D.J., “Control, and Simulation of a Dual Lift System Using Autonomous R-MAX Helicopters,” Presented at the AHS 70th Annual Forum, Montreal, Quebec, Canada, May 20–22, 2014. 11. Reuben, R., Fogel, O., Rosen, A., Berrios, M.G., and Cicolani, L.S.,”Using Wind Tunnel Tests to Investigate Dual Lift Trim, Maneuvers, Stability and Control,” Presented at the AHS 73rd Annual Forum, Fort Worth, TX May 9-11, 2017. 12. Enciu, J. and Horn, J.F.,” Flight Performance Optimization of a Multi-Lift Rotorcraft Formation,” Journal of Aircraft, Vol. 54, No. 4 July 2017. 13. Bernard, M., Kondak, K., and Hommel, G., “Load transportation system based on autonomous small size helicopters,” The Aeronautical Journal, Vol. 114, (1153), March 2010, pp. 191–198. 14. Takahashi, M.D., Whalley, M., and Schulein, G., “Flight Control Law Design and Development for an 18
Autonomous Rotorcraft,” American Helicopter Society 64th Annual Forum Proceedings, Montréal, Canada, April 29 - May 1, 2008. 15. Takahashi, M.D, Whalley, M.S., Fletcher, J., Moralez, E., Ott, C.R., Olmstead, M.G., Goerzen, C.L., Schulein, G.J., "Development and Flight Testing of a Flight Control Law for Autonomous Operations Research on the RASCAL JUH-60A," Journal of the American Helicopter Society, Vol. 59, No. 3, July, 2014. 16. Takahashi, M., Whalley, M., Goerzen, C., Mansur, H., Ott, C., Minor, J., Morford, Z., and Schulein, G., “Autonomous Rotorcraft Flight Control with Multi-Level Pilot Interaction in Hover and Forward Flight,” Journal of the American Helicopter Society, Vol. 62, No. 3, July, 2017 17. Berrios, M.G., Takahashi, M.D., Whalley, M.S., Schulein, G.J., Cicolani, L.S., and Powell, J.D. "Load Distribution Control for a Dual Lift System using RMAX UAVs with Flight Test Results," Presented at the AHS International 74th Annual Forum, Phoenix, AZ, May 14-17, 2018. 18. Anon., “Yamaha RMAX: Operations Manual,” Yamaha Corporation doc. L15-28199-01, 2001 19. Ivler, C. M., Powell, J. D., Tischler, M. B., Fletcher, J. W., and Ott, C., “Design and Flight Test of a Cable Angle/Rate Feedback Flight Control System for the RASCAL JUH-60 Helicopter,” 68th Annual Forum of the American Helicopter Society, May 2012. 20. Takahashi, M.D., Whalley, M.S., Ott, C, Minor, J.S., and Mansur, M.R.H., "Flight Test of Advanced Optical Systems (AOS) HERMES+ System for Automated Sling-load Pickup and Delivery," U.S. Army RDECOM TR RDMR-AD-15-01, June 2015. 21. Tischler, M. B. and Remple, R, Aircraft and Rotorcraft System Identification, Engineering Methods with Flight Test Examples, AIAA Education Series, American Institute of Aeronautics and Astronautics, Reston, VA, 2006. 22. Lightbody, Simon, and Chisholm, Gary,”Techniques in Relative RTK GNSS Positioning,” White paper, Trimble Marine Division, Colorado, 2010.
