六自由度动态线驱动坐标架中速度场、速度矢量场检测方案(粒子图像测速)

智能文字提取功能测试中

aerospace 2 of 16Aerospace 2016，3， 11 Aerospace 2016， 3，11； doi：10.3390/aerospace3020011www.mdpi.com/journal/aerospace Article A Six Degrees of Freedom DynamicWire-Driven Traverse Thomas J. Lambert， Bojan Vukasinovic * and Ari Glezer Woodruff School of Mechanical Engineering， Georgia Institute of Technology， Atlanta， GA 30332-0405， USA；tlambert6@gatech.edu (T.J.L.)； ari.glezer@me.gatech.edu (A.G.) * Correspondence： bojan.vukasinovic@me.gatech.edu； Tel.： +1-404-385-1889 Academic Editor： Raffaello Mariani Received： 3 February 2016； Accepted： 5 April 2016； Published： 14 April 2016 Abstract： A novel support mechanism for a wind tunnel model is designed， built， and demonstratedon an aerodynamic platform undergoing dynamic maneuvers， tested with periodic motions up to20 Hz. The platform is supported by a 6-DOF (six degrees of freedom) traverse that utilizes eightthin wires， each mounted to a servo motor with an in-line load cell to accurately monitor or controlthe platform motion and force responses. The system is designed such that simultaneous control ofthe servo motors effects motion within ±50 mm translations， ±15° pitch，±9° yaw， and ±8°roll atlower frequencies. The traverse tracks a desired trajectory and resolves the induced forces on theplatform at 1 kHz. The effected motion of the platform is measured at 0.6 kHz with a motion capturesystem， which utilizes six near-infrared (NIR) cameras for full spatial and temporal resolution of theplatform motion， which is used for feedback control. The traverse allows different platform modelgeometries to be tested， and the present work demonstrates its capabilities on an axisymmetric bluffbody. Programmable timed outputs are synchronized relative to the model motion and can be usedfor triggering external systems and processes. In the present study， particle image velocimetry (PIV)is used to characterize the realized wakes of the platform undergoing canonical motions that areeffected by this new wind tunnel traverse. Keywords： model support； 6-DOF； wind tunnel testing； aerodynamics； wakes 1. Introduction An inherent difficulty with wind tunnel investigations of nominally “free" aerodynamic bodies isrelated to their mounting in the tunnel test section. Ideally， the model support should cause little orno aerodynamic interference (such as a magnetic-force support [1，2])， but most conventional supportsystems have relied on some form of a rigid connection that can interfere with the flow around thebody and especially in its wake. The most common model support in wind tunnels is of a sting type(e.g.， Achenbach [3]) that is typically connected to either aft or a side of the model. Another relatedsupport approach is by utilization of struts (e.g.， Taylor et al. [4，5])， while the least common is a wiresupport (e.g.， Griffin et al.[6])， which is usually restricted to low flow speeds. In a case where a modelis desired to realize multiple degrees of freedom motion， a combination of sting and strut supports areoften needed (e.g.， Pattinson et al. [7]). Furthermore， there is a prevailing interest in model supportinterference with the "true" flow dynamics for transonic and supersonic flow regimes (see an extensivebibliography presented by Tuttle and Gloss [8]). This interference is primarily caused by the physicalpresence of a model support within the model wake， preventing its full development， and the shedvortices off the support structure that may interfere with the vortices naturally shed off the model.These problems have been recognized from early stages of wind tunnel testing (e.g.， Bacon [9])， anddifferent methods for either calibration or correction for the support interference have been proposedsince about a hundred years ago. As can be expected， support interference problems become exacerbated if the model support isto impose dynamic motion on the model. An extensive review of support interference in dynamicwind tunnel tests was presented by Ericsson and Reding [10]， outlining a practical guideline forbest practices in minimization of the support interference under specific flow regimes and modelgeometries. Beyers [11] pointed to the coupled interference of the support structure at the tunnelwalls in the dynamic testing that precludes existence of any analytical corrections. He suggested thatdynamic interference could be effectively eliminated in a case of rotary balance testing. The current study extends the methodology for a thin wire support， aimed at minimizing flowinterference with a traverse that can affect dynamic motion of a wind tunnel model at low flowspeeds， and is primarily influenced by earlier studies of Abramson et al. [12，13] and Lambert et al. [14].The present work investigates a wire-mounted body on a programmable six degrees of freedom(6-DOF) (x/y/z-translation and roll/pitch/yaw) eight-wire traverse that is electromechanically drivenby a dedicated feedback controller to remove the parasitic mass and inertia of the dynamic supportsystem and of the model. This traverse is designed to support any model with eight mounting points，aimed at estimating the aerodynamic force that the model without any support would experience，given： (1) The flow is not in the transitional speed range， where the wires’ wake could alter theflow's transition to turbulence； and (2) The minimum length scale of the model is at least an orderof magnitude larger than the diameter of the support wires to ensure decoupling of their respectivevortex shedding. For demonstration， an axisymmetric body is used for the purpose of illustrating the controlauthority of the current developed traverse system. 2. ExperimentalMethodologies 2.1. Physical Traverse Components This traverse is designed as an add on to an open-return wind tunnel at Georgia Tech having atest section that measures 91 cm× 91 cm and a maximum speed of 40 m/s， and is also designed suchthat it can be incorporated and removed from the tunnel， if desired. The support mechanism usesmultiple strands of wire rope (aircraft cables) to suspend an aerodynamic platform inside the windtunnel， and is shown in Figure 1a. The wire ropes have a very small cross-sectional area of 1.13 mm-and they have low blockage， thus minimizing local acceleration of the airstreams in the vicinity ofthe model. Theoretically， only six wire ropes would be required for a statically-determinate supportof the model. However， the geometry of the wind tunnel and the desire to keep the wire rope andactuators near the corners of the tunnel cross-section led towards an over-constrained support byeight wire ropes. The end of each wire rope is fastened to a servo actuator (YASKAWA SGMAV-10A，having peak torque of 9.55 N. m at a 6000 rpm peak)， through a driving pulley. Each servo actuatoris mounted outside of the tunnel and dynamically feeds a wire rope inward or outward to alter theposition and orientation of the suspended aerodynamic platform. The corner flanges of the windtunnel are modified to provide eight small openings with enough clearance to accommodate thedesired range of wire motion and therefore model motion. It is noted that the motor locations arespaced vertically by 120 cm， horizontally by 128 cm， and streamwise by 176 cm， where these locationsare chosen for convenience and can be optimized for future performance objectives， if necessary. All ofthe driving pulleys have a secondary attachment to an extension spring (with a constant of 208.4 N/m)that provides a preload tension on each wire when the model is in equilibrium (for the current casestudy， the preload is approximately 60 N). The tension in each mounting wire is measured using aninline miniature load cell (FUTEK FSH00097， with a range of 0-110 N， a non-repeatability of 0.055 N，and a mass of 0.02 kg). The temporal response of the raw load cell signals is on the order of 10 kHz， butthese signals are pre-filtered and sampled at a rate of 1 kHz to match the controller rate (see Section 2.4).Figure 1b shows one constructed servo actuator assembly， with labeled components， as all eight servoactuator assemblies are identical. Figure 1. Computer-aided design (CAD) model of the 6-DOF (six degrees of freedom) traverse (a)； anda model of a servo motor assembly (b). During the system design phase， several different types of motor actuators were consideredfor this application. Gear motor actuators could provide more output torque than the direct-driveconfiguration， but they would have too much inertia to meet the dynamic response goals. All typesof linear actuators examined had limitations with response speed or maximum achievable wire ropetension， where the design goals for motion are 50 Hz controlled motions of model deflections ofup to 5 mm， and wire tensions of up to 110 N. A vane-type hydraulic rotary actuator would be acompact， high torque， and high response rate solution， but it would require a hydraulic reservoirand pressurization system， and it was deemed less applicable compared to the realized rotary servoactuator shown in Figure 1b， although it would also achieve all of the design goals. 2.2. Motion Capture System The primary goal of this traverse is to achieve accurate real time motion control of an aerodynamicplatform with disturbance rejection. A VICON MX Motion Capture System was acquired for measuringthe spatially- and temporally-resolved motion of the wire-mounted model in the wind tunnel (Figure 2).This system consists of： (1) six MX-T40S cameras， each with a 12.5 mm lens (capable of up to 515 fps atfull frame， and 2000 fps with a limited view)， an electronic freeze frame shutter， and a resolution of4 megapixels； (2) a MX-Giganet interface of the cameras to a camera host computer (provides power，synchronization， and data transfer)； and (3) VICON Tracker 2.0 Software for low latency data streamingto the traverse controller computer. Figure 2. Schematics (a) and images (b) of the motion analysis system camera orientation. The cameras are mounted around the wind tunnel's test section and are focused on thewire-mounted model. The current system utilizes overlapping fields of view to determine thetime-resolved (6-DOF) position of the moving model in the wind tunnel at a frequency of 600 Hz.The model's position is tracked using an optical correlation based on an array of small reflectiveround markers that are affixed to the model to reflect near infrared (NIR) light that is emitted bydiode array built into each camera. The placement of the cameras can be arbitrary as long as thesix projected images of these markers measured by the cameras contain enough information to fullydefine the 6-DOF motion of the model (shown in Figure 2a with both the front and top views ofthe system). The camera system is calibrated by dynamically moving a calibration target with threereflective markers through the measurement volume. Once the system is calibrated， a static image ofacalibration plate of reflective markers is measured to construct a template that is saved on the hostcomputer to transform the marker measurements into the dynamic aerodynamic model coordinateswith respect to the wind tunnel frame of reference. An example of the chosen camera locations areshown in Figure 2b. After the system is calibrated， the VICON Tracker software calculates the errorin each marker (having typical values between 0.05 and 0.5 mm， dependent on the calibration andcamera locations)， which are used to calculate the error in the measured model coordinates. 2.3. Case Study： Axisymmetric Model This traversing system is designed to handle a variety of models including munitions， small-scaleunmanned aerial vehicles (UAVs)， parafoils，parachutes， and rotorcraft configurations. For the purposesof testing of the traverse， the model used as an example case study is an axisymmetric body， as shownin Figure 3， which has been of interest in prior investigations (e.g.[12-14]). This axisymmetric windtunnel model (Figure 3) is assembled using stereo-lithographed (SLA) and aluminum components(D=90 mm，L=165 mm， m=0.53 kg). The mid-section and nose are hollow and connected to thetail section by a central shaft. After assembly， the SLA model surface is painted black resulting in anaverage surface roughness of approximately 5 um and then machined flush on a lathe with the centralpiece. In addition， six reflective circles of 7 mm diameter are attached on the surface for the motionanalysis system， and each marker was made from 3M 7610 retroreflective tape of 0.4 mm thickness.These markers are arbitrarily placed so that each camera sees three of them (three of which can also beseen in Figures 2b and 3c). Figure 3. Side (a) and upstream (b) views of the CAD wind tunnel model with four hybrid fluidicactuators marked in green， and the model supported by the traverse (c). A secondary requirement for this traverse is that it can provide electrical power to the windtunnel model to enable future use of flow control applications. In the axisymmetric model example，four independently-driven synthetic jet actuators (labeled the top， left， bottom， and right actuators，as shown in Figure 3b) would be， in principle， integrated and distributed along the perimeter of thetail section， as they would comprise the flow control components in a follow-up study. The electricalpower that would be used as a source for the synthetic jets is routed through weaved positive voltagelines (30 AWG) along the mounting wires on the downstream end of the model (shown in Figure 3c)，where the entire system is grounded. The range of translational and rotational motions of the traverse are heavily dependent on thelocation where each mounting wire is attached to the model (the range of translational motions are also dependent on the locations of the eight servo actuators). For example， if the wires are mountedsuch that they are equally distributed in the azimuth with 90° between each wire， a roll motion wouldbe impossible， and this scenario is shown in Figure 4a. If the wires are not equally distributed in theazimuth， two symmetric groups of wires can be tightened or loosened to provide roll， as shown in redand green in Figure 4b. The theoretical maximum roll (with full controllability) occurs when one ofthese symmetric groups becomes in line with the center of the wire mounts， where this limit case isshown in Figure 4c， which occurs if the green set of wires in Figure 4b are tightened. A similar argumentcan be applied to both the pitch and yaw directions of the model looking at the side of the model or thetop of the model， respectively. It is noted that in the front view shown in Figure 4， the projections of thefour upstream and four downstream wires overlap， and these respective overlapping wires need to betightened or loosened equally in order to prevent any pitch or yaw dynamics. The current mountinglocations were chosen such that the model can theoretically have a significant and comparable rangein all three rotational directions. These realized locations are centered on the model and displacedvertically by ±35.7 mm， horizontally by ±27.4 mm， and streamwise by ±26.3 mm. The extent of therealized traverse motion is discussed in detail in Section 3.1. Figure 4. Orientation of the support where pure roll is disabled (a) and enabled for clockwiseroll (green arrows) and counter-clockwise roll (red arrows) (b)； along with its respective maximumclockwise roll deflection (C). 2.4.Electrical Traverse Components The electrical subsystem of this traverse can be considered as being comprised of three separatecomposite elements. The first element is a "host"computer which is used for implementing the realtime traverse controller and has two Quanser Q8 real time data acquisition boards which are used forsignal communication of the servo motor command and encoder signals， as well as the flow controlactuator commands. The second element is a high power distribution system that routes up to 1 kWpower to servo amplifiers for each of the servo motors. Particular attention was given to proper wirerouting and grounding in order to minimize electromagnetic interference (EMI) contamination of themeasured signals. Each servo motor has an integrated 20 bit differential quadrature absolute encoder，where the output is converted to a single ended signal by custom electronics， and is sampled by thehost computer through one of the Q8 boards. The third element is a custom multichannel Ethernet dataacquisition (DAQ) system that samples the load cells. The load cell signals are on the order of mV andare susceptible to noise， so each load cell is routed with a short cable to signal conditioning electronicsthat have an instrumentation amplifier and passive resistor-capacitor (RC) filtering. These signals arethen transmitted through universal data protocol (UDP) packets at a nominal frequency of 1 kHz tothe host computer. The user interfaces with a master computer that uses Simulink as an interface to build the controller.Upon completion of the desired controller， Simulink generates a "C" code that is cross compiled into aQNX binary. This binary is copied across an Ethernet network to the host computer (which utilizesa QNX operating system). When the master computer sends the execution command， the controllerbinary on the host machine is executed in a hard real time environment， which sends time-stampedsignals back over Ethernet to the master computer for accurate data recording. The host machine isalso connected via Ethernet to the broadcasted signal from the VICON machine to use the position measured from the camera system for feedback to the control system code. It should be noted thatthe driving signals for flow control can also be generated by the controller and output from the hostcomputer for referencing/triggering of external waveforms and signals. 2.5. Control Systems Design The designed traverse is controlled through the trajectory-tracking controller depicted in Figure 5.First， the user can provide two command inputs； a time trace for the desired model trajectoryin 6-DOF， and a time trace for the actuation commands to， for example， amplitude modulate allfour synthetic jet carrier waveforms， when flow control needs to be utilized. Second， the 6-DOFcommanded motion is converted into eight servo motor commands (6-DOF to 8-DOF)， which iscalculated through the geometry of the model and chosen mounting points， assuming the wiresare incompressible. The command signal to each servo actuator is then generated using eightproportional-integral-derivative (PID) controllers with the same coefficients (Kpinner =4.24N·m/rad，Kl，inner =21.20 N·m/(rads)， (KD，inner=0.02 N·m·s/rad) which commands the torque in the motors.It is noted that with pure feedback control， there will be a phase lag between the desired commandand the realized command， so the motions are currently restricted to time harmonic motions whereoptimal PID coefficients yield an output that is self-similar to the desired command with a changein phase and amplitude. The realized motion is then calculated through measuring the conditionedencoder signals and inverting these servo motor positions to the model position using a least squaresalgorithm (8-DOF to 6-DOF). This control loop is considered the "inner loop controller" in Figure 5.It is noted that this inner loop controller is sensitive to errors in the geometric parameters of thetraverse (e.g.， mounting locations， or damping and friction of the motors). In order to compensate forthis geometric model， as well as the change in the amplitude of the desired command， an“outer loopcontroller"shown in Figure 5 is implemented to adjust the command of the inner loop based onthe motion analysis system to allow for accurate trajectory tracking. The outer loop uses an integralerror feedback controller that is set such that the measured trajectories reach the desired ones within10 s (KL.outer =0.3 s-1 for the 3-DOF translational commands and 0.1 s-1 for the 3-DOF rotationalcommands). In addition， a manual gain of the inner loop command is implemented， determined bymatching the magnitudes of the measured amplitude to the desired amplitude， which is dependent onthe frequency and desired motion. Figure 5. Schematics of the traverse trajectory-tracking controller. The real time wire tensions measured through the load cells are used to extract the aerodynamicforces and moments on the model. This is realized by using a free body diagram in a control volume(CV) shown in Figure 6， where the aerodynamic forces are calculated by the inertial force， F， of themodel (acting at the center of gravity， CG， and calculated from its estimated acceleration) with all otherforces subtracted. These other forces include the weight of the model (FG， acting at CG)， the tension inthe wires， Fr， and the estimated aerodynamic drag of the wires alone， FAwires. The latter is calculatedassuming the wires are cylinders with pure form drag (Cp =1.25) and no friction drag， using thereal time location of both the model and motor wire ends and calculating the normal projection ofthe flow. It is noted that each wire force is estimated as acting on the midpoint between the tunnelflange and the respective real time location of the wire mount on the model， and acting in the direction normal to the wires. The resultant force is the aerodynamic force on the model， FAmodel， and a similar approach can be used to determine the moments by weighting each force by the location at which it acts. Measurements of the aerodynamic forces for the axisymmetric body using this force balance are presented in Section 3.2. The current trajectory-tracking controller described in this section restricts the model to a prescribed trajectory， where the model has a negligible response to external aerodynamic forces (where the disturbance force is on the order of ~1 N， and the spring constant of the cables are on the order of ~100，000 N/m). It should be noted， however， that a free flight controller can instead be designed where the model would respond directly to the disturbance forces (by using the load cells as the primary feedback instead of the encoders and camera signals)， but a more aggressive controller would have to be implemented in the inner loop to induce dynamics that do not significantly lag the induced forces， and this type of controller is considered for future work. Figure 6. Free-body diagram of the aerodynamic platform and the support wires with the aerodynamicmodel (blue)， inertia (green)， gravitational (cyan)， and aerodynamic wire (orange) forces， and the wiretensions (red). 3. Results 3.1. Motion Response The system dynamic performance is characterized in terms of displacement range and frequencyof sinusoidal commands， and is depicted in Figure 7. Although the mounting points of the wires onthe motors and the model determine the ideal maximum range of motion with limitless tension andunbreakable wires， this range diminishes for a fixed maximum tension. This range is also frequencydependent and diminishes with higher frequencies as the wires have to support an accelerating modelwith a larger inertial force. Therefore， the current controller is tested with a range of frequencies andthe maximum range of motion in all six degrees of freedom is measured with a maximum wire tension of 110 N，and the results are shown for translations and rotations， in Figure 7a，b， respectively. Thisfrequency response is dependent on all of the aforementioned geometric and controls parameters andcan be further optimized by modifying these parameters， if necessary. The measured displacementamplitude that is realized is 50 mm in translation (x， y， and z)， and rotations of 8°xx， 15° ly，and 9° azat a frequency，f= 0.5 Hz， and decreases to 5 mm (x， y， and z)， 1° ux，3°y， and 2° xz at f=20 Hz. Fourof these motions are visualized in Figure 7c， with streamwise (x) translation， vertical (z) translation，pitch (cy)， and yaw (z). Figure 7. Frequency response of tested translational (a)， and rotational (b) motions， with an illustrationof the maximum traverse displacements achieved in four of the degrees of freedom (c). To test the control authority of the traverse for 6-DOF motion tracking， a combination of rotationsis commanded (pitch and yaw out of phase) and the instantaneous measurement is shown in Figure 8with streamwise translation x (Figure 8a，b)， cross stream translation y (Figure 8c，d)， vertical translationz (Figure 8e，f)， roll x (Figure 8g，h)， pitch ay (Figure 8i，j)， and yaw az (Figure 8k，l)， respectively.The commanded signal is a 3° amplitude sinusoidal pitch and sinusoidal yaw， 90° out of phasewith each other. This same motion is executed with a representative“low" frequency of 1 Hz inFigure 8a，c，e，g，i，k along with a "high" frequency of 10 Hz in Figure 8b，d，f，h，j，l. Both of these motionsare executed with wind tunnel flow (Rep=2.3×10) to demonstrate the disturbance rejection of thecontroller. The time resolved path captured by the motion analysis system is shown in blue whilethe commanded signal is shown in dotted gray. These data show significant agreement between thecommanded and measured signals in terms of peak-peak magnitude (within 5% at 1 Hz in Figure 8i，kand 7% at 10 Hz in Figure 8j，l)， with a phase lag of 40° at 1 Hz and 30° at 10 Hz. The realized deviationsin the four other degrees of freedom that are not commanded are small： within 0.5° in roll (Figure 8g，h)，and within 0.3 mm in all translations (Figure 8a-f). As mentioned in Section 2.2， the errors in themotion measurements for this data set (which is the same as for the motions in Sections 3.2 and 3.3) arecalculated using the motion analysis system calibration error output (0.45 mm per marker on average)，yielding errors of 0.18 mm in translation and 0.02° in rotation. It is noted that these instantaneoustraces are measured 10 s after the traverse is activated after the transient response of the traverse hassettled. The measurements that are shown in Figure 8 can also be used as feedback to the triggeringlines for flow diagnostics (e.g.， particle image velocimetry (PIV)) and/or flow control， and therefore aslong as the flow control is a function of the measured DOF the phase lag between the commanded andmeasured trajectory is inconsequential for the considered harmonic motions. In this section， the wind speed is fixed at 40 m/s and it is noted that the traverse controller treats the external aerodynamicforces on the model as disturbances to be rejected with PID control. Keeping fixed PID coefficients， thephase lag of the commanded and realized responses would increase with increasing wind speed， asthe disturbance force grows， but in the present investigations， the PID parameters are pre-adjusted fora given Rep and can be further tuned with a change in the wind speed. Figure 8. Traverse motion response for 3° amplitude pitch and yaw rotation， 90° out of phase， withmeasured translation in x (a，b)， y(c，d)， and z(e，f)， and rotations in roll (g，h)， pitch(ij)， and yaw (k，l)，for frequencies of 1 Hz (a，c，e，g，i，k) and 10Hz (b，d，f，h，j，l)， with a flow Rep=2.3×105. The commandedand measured responses are shown in dotted gray and blue， respectively. 3.2. Force Response The second analysis that is implemented on the traverse is a validation of the force and momentmeasurements. Initially， the drag force on the stationary centered model (Fp) is measured over a rangeof wind tunnel speeds， Uo， of up to 40 m/s (0 关闭