Unmanned aircraft design, development and
applications at the Unmanned Systems Lab
Kevin Kochersberger
Mechanical Engineer...
Outline
Lab overview
Morphing wing aircraft
Precision agriculture
Image-based 3D reconstruction
Theory
Applications
Roboti...
Brief overview of the lab
The Lab consists of a 3,000 sq. ft. of high
bay area, graduate student office space,
and mechani...
The Unmanned Systems Lab
The Lab’s focus is in autonomous aircraft
Sensing payloads carried by fixed and rotary
wing aircr...
Wing morphing is based on
piezoelectric actuator technology
Macro Fiber Composite (MFC) actuators show large strains
compa...
Morphing concept
In a bi-morph configuration, substantial bending
deflection can be obtained with the MFC actuators
This b...
The original morphing aircraft
design proved the concept
The 2007 aircraft design was originally created to explore
both p...
A study was conducted to determine
better morphing configurations
2-D wind tunnel testing was conducted on a variety of mo...
Top and bottom bimorphs (4
actuators) allow a CL range > 1.2
9
2-D wind tunnel results: higher delta
CL and lower drag with morphing
10
Hysteresis Modeling
Inverse Hysteresis Operator (IHOp) used to provide feed-forward
control for MFC operation
It basically...
High voltage conversion
Custom electronics used to power the MFC actuators were developed via a
Phase II SBIR with AVID LL...
Full wing morphing
UAV applications in targeting can benefit from high bandwidth,
agile handling
Servo driven ailerons lac...
Full wing morphing
Entire wing planform morphs instead of discrete control surfaces
Simulations in FEA performed to determ...
Flight Comparison: Roll
Coefficient
Measure Roll Rate: 80% Amp, 0.2s Duration
MFC roll coefficients from
simulation verif...
Frequency Response:
MFC vs. Servo
From Ohanian, O, David, B., Taylor,
S., Kochersberger, K., Probst, T.,
Gelhausen, P., C...
Swashplate-less rotorcraft
flight control
Use blade torsional harmonic
excitation via MFC to phasecontrol asymmetric lift...
UAS for the Monitoring of Crop Status and Stress
Donnie Rogers
Graduate Research Assistant
Virginia Tech – Mechanical Eng...
Overview of ag work
The literature review focused on three main topics as they relate
to the crop monitoring project:
1. S...
Imaging Payload
The payload developed for the project uses a multi-spectral
camera to gather images in both the Visible an...
Flight tests at Kentland Farms
The purpose of the flight was to determine the applicability
of unmanned aircraft for crop...
Ground Image Segmentation &
Results
A two stage segmentation process was used to isolate the corn from
background, and the...
Statistical Results
A Least Significant Difference (LSD) test has shown that three of the four
Nitrogen treatments are det...
Grain Yield Correlations
Grain yield totals were measured for each N rate treatment after
harvest and compared to the calc...
Tobacco Imaging Study
This series of flights performed in September on late season
tobacco were intended to explore the us...
Tobacco Survey Results
Two neighboring 8-row blocks appeared to have different levels of
growth (most likely, different hy...
Bleached Leaf Segmentation
A segmentation method was developed to isolate both the green
tobacco leaves and the bleached l...
Leaf Segmentation Results
The leaf segmentation method was successful in isolating both
groups of tobacco leaves from each...
UV and LWIR Imaging Results
Data Processing:
– UV and IR JPEGS appear to have been distorted by motion blur caused
by heli...
Image-based 3D reconstruction
•
Image-based 3D reconstruction is ideal for robotics
•
Cameras are lightweight payloads
•
...
Stereovision Basics
Stereovision is 3D mapping technique that utilizes two or more
cameras, which are separated by some di...
Stereovision Basics
The vertical resolution of the 3D points is directly related to the
distance between the cameras and t...
Application
• A 5’ baseline system is currently
used for aerial imagery on the
RMAX
• Methods developed to correct for
vib...
Stereo examples
34
Structure-from-Motion (SfM)
methods
Bundler SfM output
PMVS output
Bundler and PMVS output
Improved accuracy over ster...
Some examples
Stockpile mapping
Volume estimation
Structural defect detection
36
Anomaly detection in roadway via
aerial imagery and 3D reconstruction
Three disturbed areas were successfully identified a...
UAV path planning for ground
vehicles
Area traversal of ground vehicles could
be greatly improved if coupled with a
UAV im...
3-D Scene Understanding
Challenge: classify 3D environments efficiently for
damage assessment, threat assessment,
ecosyste...
3-D Scene Understanding
The lab is using a hexacopter to carry a 48” wide baseline imaging
system to collect nadir and sid...
Training and classification
Use of k-means clustering breaks the scene
into 3D “super voxels” based on location only
Furth...
The resulting classification
has many uses
Provides spatially relevant data to support other analyses
Ecosystem health
m...
Organic Disaster Response
Mid-level damage assessment can be facilitated with a 90 kg – class
autonomous helicopter
1
2
...
Flight Operations
A mission architecture based on a 90 kg helicopter / 18 kg payload is
designed to maximize information w...
Radiation detection capabilities
• Oct, 2009 test demonstrated detector
function with a single, collimated
source
• June, ...
Radiation detection mission
June, 2010 test: Successfully mapped single
and multiple omni-sources at Savannah River
Nation...
Use of spatially variant deconvolution
for source localization
Radiation intensity measurements are correlated with height...
Radiation detection mission
An optional radiation localization flight can be performed to
more accurately locate a source ...
Ground sample collection
robot
A helicopter-deployed ground sampling robot was developed to retrieve
radioactive samples
B...
Remote ground robotic
operations using a virtual display
Map 3D terrain
with vision system
Plan traversability
path usin...
Intelligent radio repeating
using mapped terrain and A*
Note: The blue lines denote the strongest radio link. Notice how ...
Intelligent radio repeating
using mapped terrain and A*
Note: The blue lines denote the strongest radio link. Notice how ...
Test data – Intelligent radio
repeating
53
Conclusions
The positive uses of UA will be adopted by
a suspicious public (as all promising
technologies have been adopt...
Questions?
of 55

Autonomy Incubator Seminar Series: Unmanned aircraft design, development and applications at VT's Unmanned Systems Lab

Autonomy Incubator Seminar Series: Unmanned aircraft design, development and applications at the Unmanned Systems Lab presented to NASA LaRC by Dr. Kevin Kochersberger, Virginia Tech on 16 December 2013
Published on: Mar 3, 2016
Published in: Technology      Business      
Source: www.slideshare.net


Transcripts - Autonomy Incubator Seminar Series: Unmanned aircraft design, development and applications at VT's Unmanned Systems Lab

  • 1. Unmanned aircraft design, development and applications at the Unmanned Systems Lab Kevin Kochersberger Mechanical Engineering kbk@vt.edu 540-231-5589 1
  • 2. Outline Lab overview Morphing wing aircraft Precision agriculture Image-based 3D reconstruction Theory Applications Robotic collection IED-detect and path planning Terrain feature mapping and classification Organic VTOL disaster response 2
  • 3. Brief overview of the lab The Lab consists of a 3,000 sq. ft. of high bay area, graduate student office space, and mechanical and electrical shop resources 8 graduate students 30 undergraduates 1 lab technician 3
  • 4. The Unmanned Systems Lab The Lab’s focus is in autonomous aircraft Sensing payloads carried by fixed and rotary wing aircraft that weigh 1 – 90 kg Novel aircraft design Novel control strategies 4
  • 5. Wing morphing is based on piezoelectric actuator technology Macro Fiber Composite (MFC) actuators show large strains compared to competitive piezoelectric actuators Note that these operate at high voltages ~ 1500 V Maximum Strain-Stress Outputs from several active materials. [Williams] 5
  • 6. Morphing concept In a bi-morph configuration, substantial bending deflection can be obtained with the MFC actuators This bending behavior is suitable for aerodynamic control Positive Actuation Negative Actuation 6
  • 7. The original morphing aircraft design proved the concept The 2007 aircraft design was originally created to explore both pitch and roll control using MFC technology Goal was to achieve high efficiency and high bandwidth It flew, but barely! 7
  • 8. A study was conducted to determine better morphing configurations 2-D wind tunnel testing was conducted on a variety of morphing configurations to find an optimal design One thick wing design is a thin morphing surface with a bottom wiper that maintains contact with the top surface 8
  • 9. Top and bottom bimorphs (4 actuators) allow a CL range > 1.2 9
  • 10. 2-D wind tunnel results: higher delta CL and lower drag with morphing 10
  • 11. Hysteresis Modeling Inverse Hysteresis Operator (IHOp) used to provide feed-forward control for MFC operation It basically curve fits the forward and backward linear functions to compensate for MFC behavior Without IHOp the pilot has a difficult time controlling the aircraft Command 11
  • 12. High voltage conversion Custom electronics used to power the MFC actuators were developed via a Phase II SBIR with AVID LLC The high-voltage system is designed to power 2 bi-morphs Command IHOp compensation 2-bimorph HV converter 12
  • 13. Full wing morphing UAV applications in targeting can benefit from high bandwidth, agile handling Servo driven ailerons lack reliability and can be difficult to integrate into a thin wing What if we could simultaneously achieve camber change and a delta angle of attack to achieve higher roll rates? Conventional controls Wing morphing achieved with piezoceramic actuators 13
  • 14. Full wing morphing Entire wing planform morphs instead of discrete control surfaces Simulations in FEA performed to determine best configurations Performance confirmed in flight test with roll doublets 14
  • 15. Flight Comparison: Roll Coefficient Measure Roll Rate: 80% Amp, 0.2s Duration MFC roll coefficients from simulation verified in flight test: 300 P(deg/s) 200 100 0 Flight: Simulation: -100 -200 -300 0 0.2 0.4 0.6 1 0.8 1.2 1.4 Croll = 0.036 Croll = 0.055 1.6 t(s) input signal 100 Servo MFC MFC Thin Wing 50 0 -50 -100 0 0.2 0.4 0.6 1 0.8 1.2 1.4 1.6 1.8 t(s) From Ohanian, O, David, B., Taylor, S., Kochersberger, K., Probst, T., Gelhausen, P., Climer, J., “Piezoelectric Morphing vs Servo-Actuated MAV Control Surfaces, Part II: Flight Testing,” to be presented at the AIAA ASM 2013 Conference, Grapevine, TX, Jan 9, 2013. Images courtesy of AVID, LLC 15
  • 16. Frequency Response: MFC vs. Servo From Ohanian, O, David, B., Taylor, S., Kochersberger, K., Probst, T., Gelhausen, P., Climer, J., “Piezoelectric Morphing vs ServoActuated MAV Control Surfaces, Part II: Flight Testing,” to be presented at the AIAA ASM 2013 Conference, Grapevine, TX, Jan 9, 2013. Images courtesy of AVID, LLC Dramatic increase in bandwidth with MFC MFC has small phase lag in comparison to servo-driven controls 16
  • 17. Swashplate-less rotorcraft flight control Use blade torsional harmonic excitation via MFC to phasecontrol asymmetric lift and produce roll and pitch moments The blade torsional frequency must be tuned to the 1 per rev frequency (Of course) It allows trimming in forward flight
  • 18. UAS for the Monitoring of Crop Status and Stress Donnie Rogers Graduate Research Assistant Virginia Tech – Mechanical Engineering Unmanned Systems Lab 18 Copyright 2013, Donald Rogers
  • 19. Overview of ag work The literature review focused on three main topics as they relate to the crop monitoring project: 1. Spectral Reflectance of Vegetation  Reflectance data is a primary tool for agronomic remote sensing.  Led to the formation of ratios of reflectances, aka vegetation indexes. Most commonly NDVI.  NDVI suffers from soil background interference. RNIR = NIR reflectance, RRed = Red reflectance 2. Advantages of UAS in Crop Monitoring 3. Remote Sensing of Corn and  Low altitude flights results in high resolution Tobacco data.  Can collect data more frequently than manned aircraft or satellite.  Affordable and multi-purpose.   Strong correlation between NIR and Green reflectance for Corn nitrogen fertilizer status. Strong correlation between NDVI and tobacco nutrient status. 19 Copyright 2013, Donald Rogers
  • 20. Imaging Payload The payload developed for the project uses a multi-spectral camera to gather images in both the Visible and Near-Infrared bands. Payload Visible NIR On-board Computer Camera (JAI AD080 GE) Switch Images from both spectrums will be combined with software to estimate crop health. 20
  • 21. Flight tests at Kentland Farms The purpose of the flight was to determine the applicability of unmanned aircraft for crop health monitoring Corn was chosen as the crop for initial testing Yamaha RMAX carries EO and NIR cameras 21
  • 22. Ground Image Segmentation & Results A two stage segmentation process was used to isolate the corn from background, and then NDVI was used to examine the N effect on yield GNDVI Mask NDVI Final Image 22 Copyright 2013, Donald Rogers
  • 23. Statistical Results A Least Significant Difference (LSD) test has shown that three of the four Nitrogen treatments are detectable by the system. 170 Mean Pixel Intensity (MPI) 160 150 Each treatment was sampled 90 times from the flight images. 140 130 The LSD test was conducted with a 5% margin of error. 120 110 100 90 N Treatment Means Best Fit Curve, R2=0.97 80 70 50 100 150 200 Nitrogen Rate (kg/ha) 250 Our testing has shown diminishing returns as Nitrogen Rate exceeds 150 kg/ha 23
  • 24. Grain Yield Correlations Grain yield totals were measured for each N rate treatment after harvest and compared to the calculated average NDVI. A correlation is observed with NDVI suggesting yield estimation is possible for HR hybrid Weaker correlation with AM hybrid due to on outlier The system has the potential for estimating grain yield 24 Copyright 2013, Donald Rogers
  • 25. Tobacco Imaging Study This series of flights performed in September on late season tobacco were intended to explore the use of several sensors: Ultraviolet (300 – 400 nm) Visible (400 – 700 nm) Near Infrared (700 – 1000 nm) Long wave IR (8000 – 12,000 nm) Flight tests were in partnership with American Aerospace Advisors, Inc. RMAX flown near South Hill, VA under COA authorization 25 Copyright 2013, Donald Rogers
  • 26. Tobacco Survey Results Two neighboring 8-row blocks appeared to have different levels of growth (most likely, different hybrids) NDVI for the 1st 8 rows = 0.209 NDVI for the 2nd 8 rows = 0.195 Although the NDVI levels are overall low (as expected due to a large number of bleached leaves), the results still show a significant difference. Visible NDVI 26
  • 27. Bleached Leaf Segmentation A segmentation method was developed to isolate both the green tobacco leaves and the bleached leaves from background noise. Segmentation is possible using the four primary content groups in the visible image Group Green Tobacco Bleached Tobacco Bare Soil Shadowed Soil The segmentation should output two images: 1st – Green Leaves Only 2nd – Bleached Leaves Only RGB Value Distinguishing Feature 94, 95, 67 Green pixel value significantly greater than blue value 148, 156, 166 Brightest pixel values in image 58, 55, 56 Low overall pixel intensity 14, 16, 15 Low overall pixel intensity Set a Green Level Threshold Set Bright Level Threshold “Noise Floor” established The distinguishing features of each group make simultaneous segmentation possible. 27 Copyright 2013, Donald Rogers
  • 28. Leaf Segmentation Results The leaf segmentation method was successful in isolating both groups of tobacco leaves from each other and background noise. Original Image Green Leaves Only Bleached Leaves Only 200628 pixels 49362 pixels Trial segmentation results provided us with a estimate of bleached leaf concentration – useful to determine time of harvest 28 Copyright 2013, Donald Rogers
  • 29. UV and LWIR Imaging Results Data Processing: – UV and IR JPEGS appear to have been distorted by motion blur caused by helicopter vibration – Individual visible wavelength JPEGS were clear, but adjacent sequential frames showed differences in magnification and orientation which affects stitching Ultraviolet Visible 2013 American Aerospace Advisors Inc. All Rights Reserved Long Wave IR 29
  • 30. Image-based 3D reconstruction • Image-based 3D reconstruction is ideal for robotics • Cameras are lightweight payloads • Cameras can be easily attached to robotic platforms (e.g. quadrotors) • Several vision techniques can process the images • Reconstructions include colors and camera locations Kentland farm equipment shed PMVS meshed model 30
  • 31. Stereovision Basics Stereovision is 3D mapping technique that utilizes two or more cameras, which are separated by some distance, to determine point correspondences in 3D. Image from Bradski and Kaehler 31
  • 32. Stereovision Basics The vertical resolution of the 3D points is directly related to the distance between the cameras and the distance from the cameras to the observed objects. Resolution for 1.5 m baseline and focal length of 8mm for 1600x1200images at the center of image.(X = Y = 0) While geo-referencing a single pair of images is easy, area scans pose a challenge 32
  • 33. Application • A 5’ baseline system is currently used for aerial imagery on the RMAX • Methods developed to correct for vibration-induced errors • Images are acquired with 80% overlap to generate 3-D terrain mosaics 33
  • 34. Stereo examples 34
  • 35. Structure-from-Motion (SfM) methods Bundler SfM output PMVS output Bundler and PMVS output Improved accuracy over stereovision 35
  • 36. Some examples Stockpile mapping Volume estimation Structural defect detection 36
  • 37. Anomaly detection in roadway via aerial imagery and 3D reconstruction Three disturbed areas were successfully identified and mapped using the Unmanned Systems Lab Yamaha RMAX helicopter equipped with the USL dual camera imaging boom. The gravel piles ranged in height from 3” to 10”. Two sets of 3D reconstructions were used for anomaly detection. 9’ Gravel piles detected as anomalies 37
  • 38. UAV path planning for ground vehicles Area traversal of ground vehicles could be greatly improved if coupled with a UAV imaging system Weighted A* or RRT* methods are applied to 3D maps from aerial imagery to find optimal paths in unstructured environments Roadway detection algorithms used with learning methods to find obstructions for optimal routing Mosaic map K-means segmentation Pre-erosion Final connected K-means paths membership 38
  • 39. 3-D Scene Understanding Challenge: classify 3D environments efficiently for damage assessment, threat assessment, ecosystem monitoring, etc Original research is focusing on urban environment classification, however any environment of interest can be used to train for classification: Hurricane damage Tornado blowdown Vehicle disruption 39
  • 40. 3-D Scene Understanding The lab is using a hexacopter to carry a 48” wide baseline imaging system to collect nadir and side view imagery Performance metric Value Endurance Weight (with camera boom) Max speed Autopilot Arduino Stereo broom description Info Cameras Canon A810 Weight Max frame rate Max resolution Control Arduino 40
  • 41. Training and classification Use of k-means clustering breaks the scene into 3D “super voxels” based on location only Further classification occurs on the super voxels for point-ness, linear-ness and surface-ness Height is added as a feature A conditional random field (CRF) is used as the framework for training Supervoxels are then grouped from training into the following classes: Poles Ground Buildings Vehicles 41
  • 42. The resulting classification has many uses Provides spatially relevant data to support other analyses Ecosystem health monitoring Number of damaged buildings, roads post-hurricane 42
  • 43. Organic Disaster Response Mid-level damage assessment can be facilitated with a 90 kg – class autonomous helicopter 1 2 Uncertain terrain or topology 3 Unstable structures 4 4 Blocked ingress and egress routes Health concerns (particulates) 5 3 1 5 2 3 Radiation 43
  • 44. Flight Operations A mission architecture based on a 90 kg helicopter / 18 kg payload is designed to maximize information while keeping flights to a minimum Mapping 1 • High Altitude Image Capture Flight • Flight of total target area used to collect high level information • 80 meter flight altitude Mapping 2 • Combined Low Altitude Image Capture and Radiation Scanning Flight • Flight of specific target area used to collect more accurate information • 40 meter flight altitude • Also gathers radiation spectra and gross gamma ray event counts Localizing • Radiation Source Localization Flight • Constant radiation intensity following to find radiation contours • 40 meter flight altitude • Post-processing to determine best estimate of radiation source location Sampling • Sample Collection Mission • Lower Ground Sampling Robot into point of interest • Use Tele-operated Ground Sampling Robot to locate and collect samples • Retract robot and return to base for sample analysis = Optional 6 44
  • 45. Radiation detection capabilities • Oct, 2009 test demonstrated detector function with a single, collimated source • June, 2010 test with uncollimated sources GPS coordinates, rad counts and spectral data transmitted real-time to the ground control station Scan Path 37.1974 37.1972 Latitude (deg) Background-Subtracted Gamma Spectrum 600 500 Channel Counts • 400 300 37.197 37.1968 37.1966 37.1964 37.1962 200 37.196 100 37.1958 0 0 100 200 300 400 500 600 Channel No. 700 800 900 1000 -80.582 -80.5815 -80.581 -80.5805 -80.58 Longitude (deg) 45
  • 46. Radiation detection mission June, 2010 test: Successfully mapped single and multiple omni-sources at Savannah River National Labs at 40 m and 60 m AGL The RMAX flew 6 hours in three days of mapping in 98º F heat – only a single software update had to be made during the radiation mapping missions which resulted in a 20 minute delay Single and dual radiation source maps (Sandia NL) 46
  • 47. Use of spatially variant deconvolution for source localization Radiation intensity measurements are correlated with height above ground using a laser rangefinder Maximum likelihood estimation (MLE) used in deconvolution The ability to resolve close proximity strong and weak sources was demonstrated in flight tests 0.85 Ci Ir source and a 0.03 Ci Co source intensity plots. The sources are spaced 20 m apart 47
  • 48. Radiation detection mission An optional radiation localization flight can be performed to more accurately locate a source of radiation PID-implemented contour following to localize sources Particle filter method used to localize a single source 12 48
  • 49. Ground sample collection robot A helicopter-deployed ground sampling robot was developed to retrieve radioactive samples Both chunk and particulate samples have been considered for collection The system (robot and winch) weigh 10 kg Tether deployment from the helicopter allows pinpoint delivery of sampling assets + = 49
  • 50. Remote ground robotic operations using a virtual display Map 3D terrain with vision system Plan traversability path using A* Track robot Robot tracking occurs by referencing the robot to features which are mapped to 3D terrain Teleoperator sees an accurate virtual 3D environment during operation 50
  • 51. Intelligent radio repeating using mapped terrain and A* Note: The blue lines denote the strongest radio link. Notice how the best link is the radio repeating link between the ground robot, helicopter, and ground robot. 51
  • 52. Intelligent radio repeating using mapped terrain and A* Note: The blue lines denote the strongest radio link. Notice how the best link is the radio repeating link between the ground robot, helicopter, and ground robot. 52
  • 53. Test data – Intelligent radio repeating 53
  • 54. Conclusions The positive uses of UA will be adopted by a suspicious public (as all promising technologies have been adopted in the past) UA toys will outrun any other attempt at promotion of the technology UA have become the future of aviation and can co-exist and support manned flight operations
  • 55. Questions?

Related Documents