Our Team

• Duke University began competing in XPRIZEs in 2013
• Advisors:
• Martin Brooke, Ph.D,
• Douglas Nowacek, Ph.D
• Tyler Bletsch, Ph.D.
• From 7 students to over 40+
• Electrical and Computer Engineering, Mechanical Engineering, Computer Science 

​

​

Overview

• First Place $4M
• A map of 500 km2 of ocean floor
• Second Place $1M
• $1M NOAA prize data for tracking chemical signal
• 10 teams going to round two each receive $100K

Structure: Heavy Lift UAV

Initial UAV Design

Current UAV Design

Breakdown: Heavy Lift UAV

• 3 Hex Systems

Outer co-axial hex

Inner hex

• Triply redundant
• Hybrid gas–electric

Breakdown: Pod

Variable buoyancy system

IMU

SONAR

• Synthetic Aperture SONAR
• IMU for position and movement data 
• WiFi for component communication 
• Gas-based variable buoyancy system
• WiFi and LoRa® between pod and UAV for data redundancy and speed
• GPS location beacon

Breakdown: Post-Processing

SONAR readings taken in a downward circular motion

Synthetic Aperture Sonar

Motion compensation methods used to correctly reconstruct images

Cloud computing to be used for parallel processing

UAV Teams

Key Functional Requirements - Drone Body

• Design Flaws of Original Drone:
• 3D printed joints
• Added weight of centerplate
• Shape of landing base
• Primary considerations for the body included:
• Weight
• Durability
• Structural stability
• Load capacity

​

Original Design

Drone Design - Upper Body

• Chosen to be a carbon fiber hexagram
• 18 rotor design - two rotors on each of the points of the star and one rotor on each of the points of the inner hexagon
• 3D printed joints at each of the points of the inner hexagon to attach landing apparatus
• 3D printed hexagonal electronics plate mounted at center

Drone Design - Landing Apparatus

• Joints were angled and printed in order to avoid direct stress in the direction of the print planes
• Cross-supported carbon fiber three leg design
• Three triangular legs evenly spaced around the inner hexagon and angled away from the center of the drone
• Horizontal support struts between adjacent legs
• 3D printed spherical ground contacts (feet) to self-correct during landings

Final Drone Body Assembly

FEA Analysis - Drone Landing Apparatus

• Inner Hexagram Joints
• Drone weight applied to legs without cross support is likely to cause failure, especially considering structural flaws of 3D printed joints.
• Cross support distributes catastrophic load throughout the structure, resulting in a frame that has desirable failure resistance and low weight.

Drone Flight - 4/12/16

Team Hybrid

• To maximize the feasibility of this concept, the drone power system has several objectives:
• Maximize flight time (>1hr)
• Create a reliable and robust powertrain
• Produce enough power to lift 20 lbs of payload
• To meet the objectives, a liquid fuel engine is required.
• High energy to weight ratio
• Enables long flight times even with large payloads

​

Hybrid Powertrain

Key Functional Requirements - Hybrid Powertrain

• MG set must be able to recharge drone batteries
• Reliable coupling method for the motor-generator (MG) set
• 56 in-lbs of torque
• 8,500 RPM
• Sturdy and lightweight frame and mounting system for the MG set that can be mounted onto the drone body
• Must have low deflections to maintain shaft alignment
• Characterize the power band for the MG set
• Determine optimal operating conditions

Overview of the Hybrid Power System

• Brushless DC motor/generator coupled to a DLE-60 two stroke gasoline engine
• LiPo batteries supply power to the motor until the engine starts, then Hybrid system acts as a generator to power motors and recharges the batteries
• Current flows through a rectifier circuit to convert from AC to DC, then flows to the load (motors and battery for actual drone operation, and water heater elements for testing)

Powertrain - Coupling

• A new 16mm steel shaft was machined to replace the nut
• The lovejoy coupling promptly broke

​

• Latest design: two 16mm shafts bolted to the engine and generator and coupled with a 2-7/64” Lovejoy

​

Powertrain - Frame

• The inherited frame was heavy, flimsy, and allowed for MG misalignment
• A new frame was designed in CAD, machined and assembled in 6061 Aluminum with 0.19” end plates.
• The new frame is rigid, lightweight and easily handles the load.
• Deflection is kept to a minimum
• Vibration is low at operating speeds

FEA Analysis

Deflection Analysis

Frequency Analysis

• Maximum deflection at peak torque (60 lb-in) is 0.287mm at the vertices of the frame
• Stress is well below failure stress

• The frame experiences resonance at 2050, 2150, 4850 and 9800 rpm
• The motor is typically run at around 8000 rpm
• Should avoid long periods of operation at resonating rpms

​

Power Output Testing

• Measured power output at various RPMs for 6 different loads
• Water heater element circuits were used as load
• A multimeter connected at the rectifier load circuit connections measured the voltage a tachometer measured the speed of the generator

Power Curve Results

• Maximum power achieved by running system at highest allowed RPM (8500) with the lowest load
• Unable to reach ‘peak’ power in the test setup due to load not heavy enough. Need 0.25 Ohm load rated for 6kW

​

Last Hybrid Test Video

At the end of the video the coupling failed after many hours of testing for all the charts above. This is being rebuilt.

We really need this for the Drone in the contest!

Custom ESC for Hybrid Generation

• Goal: A flexible, easy to program speed controller for the brushless electric motor in the hybrid system

Past Pod Teams

• Streamlined capsule-shaped pod.
• Modular design to allow for expandability and swapping of sensor modules.
• Rocket-based ascent system
• Plastic outer shell with a marine epoxy core.

Design Detail

• Modules attach using threaded aluminum rods.
• Gray shells are 3D printed plastic.
• White material is marine epoxy.
• Sensors, sonar, and rockets embedded in epoxy for waterproofing.

Manufacturing Approach

• Molds for each module 3D printed with PLA plastic.
• AeroMarine 300/21 Epoxy Resin poured in molds and sets.
• Epoxy is lightweight, waterproof, see-through, moldable into any shape, and can withstand high pressures.
• 3D printed molds not removed for ease of manufacturing.
• Molds may crush at depth from entrained air, but structural strength comes from the epoxy which will not crush.

​

Gas Generation Options

• To ascend, the pod will generate gas to fill the Gas Bell Module, making the pod less buoyant than water.
• To generate gas, two options considered:
• Releasing high pressure compressed air from a cannister at depth.
• Producing gas with a chemical reaction, e.g. a combustion reaction.
• Model rocket engines chosen to generate gas due to affordability and simplicity.

​

Rocket Module

• Based on prior work, Estes G40-7W rockets selected for gas generating ability.
• 12 rockets arranged in pod.
• 4 rockets needed to ascend from 2000m.
• Rocket nozzle filled with wax to seal fuel from water.
• Electronic ignition system under

rockets in epoxy.

• Encased in epoxy.

Gas Bell Module

• Designed to fill with water when descending, and to fill with gas when ascending from depth.
• Designed to hold 32 gallons of gas at 2000m based on calculations from previous team.
• Gap between gas bell and rocket modules allows gas to fully displace water.

Other Modules

• Other modules designed for sensors developed by ECE teams.
• Dimensions of sensor modules can change based on size of final sensors.
• Recovery system can be integrated into Pod Communication Module.
• All modules are integrated with same aluminum rod system.

Weight/Buoyancy Analysis

• Pod within weight specification
• Pod will sink when filled with water, will float when filled with gas
• If pod becomes too heavy in future (additional modules, design changes, etc.) the epoxy can be modified with glass beads to reduce its density.

F_buoyant = ⍴_fluidVg

CFD Analysis

• CFD analysis performed for both descent and ascent of pod.
• Velocities analyzed.
• Pod design will not have excessive drag nor asymmetric velocity profiles.

Electronic Ignition System Overview

Components:

• Adafruit Feather HUZZAH ESP8266 microcontroller
• Wifi-equipped
• 9 GPIO pins
• Latching Hall Effect Sensor
• Detects magnetic field
• MOSFET based firing circuit
• 16V, 35A battery
• MOSFET Gate is controlled by microcontroller output pin

Wiring Diagram

Controller Code Fundamentals

• Two modes
• Determined by state of latching hall effect sensor
• Standby Mode (default):
• Will not allow any rockets to be fired
• Primed Mode:
• 40 minutes delay before first rocket is fired
• 5-10 minute interval between successive firings
• Microcontroller is programmable
• Firing intervals can be changed

​

Testing: Rockets Underwater

• Rockets were sealed with wax and ignited in a 500 gallon tank to test system at small scale.
• Results: rockets successfully ignited underwater in 2 out of three tests.
• Failure most likely due to imperfect wax sealing and water intrusion.

Pod Testing: Marine Lab

• Pod will be tested in the near future offshore at the Duke Marine Lab.
• Will be attached to 2000m of fishing line so that it will not be lost.
• Rocket performance will be evaluated at various depths by igniting rockets at set time intervals.
• Prior to the Marine Lab test, the rocket module will be tested in a pressure chamber to examine failure methods of the wall of the rocket.

Future Work: Integration with Other Teams

• Once the other XPrize teams finish designing their components, their sensors can easily be integrated into the design.
• Design modifications may need to occur if sensors are significantly larger or smaller than expected.

​

• If weight of pod significantly changes, epoxy mix should change to reduce density.
• Implement a depth sensor to trigger rocket firings rather than a timer.

Future Work: Variable Buoyancy System

• To automate the dropping and retrieval of the pod, a variable buoyancy system will need to be implemented.

Current Pod Design

• Streamlined capsule-shaped pod.
• Plastic outer shell with a marine epoxy core.
• Buoyant winch based descent and ascent
• ​

​

Another option: Compressed Air

• In the case that the rockets do not work at depth, the team is researching compressed air strategies.
• Team plans to look into small 4500psi compressed air canister connected to a solenoid valve to control air release
• Electronic Ignition System could be adapted to control the solenoid valve.
• Pod design would essentially stay constant.

Pod Recovery: Sonar Pod Capture System

• Capture System:
• 5 arm capture claw mechanism
• 5 servo motors are attached for locking and dropping system
• Claw automatically moves up once it contacts with target object for locking

​

​

​

​

​

https://vimeo.com/209840173

Pod Recovery: Capture Claw

• Capture Target:
• Claw capture system will use a sphere target attached to sonar pod

​

​

Ardunio 5 Servo System with 6V source

Pod Recovery: Capture Claw

• Capture Claw System:
• Each of the 5 arms of the capture system has an extra joint part for dropping of the sonar pods after locking
• Two configurations of the lock and drop systems can be seen on the right

​

Pod Recovery: Precision Landing Mechanism

• Sense-Detect-Act Model
• IR-LOCK Camera senses IR Light from the Mark-One IR Beacon
• Camera Sensor + Sonar Range Finder detects the IR Beacon
• Drone switches to preinstalled precision landing mode

Pod Recovery: IR Detection and Recognition

• Detection
• Pixy Camera needs a specific firmware loaded through specific PixyMon software version: MarkOneBeta_1.0.1.hex pixymon_beta_1.0.2.exe
• Require camera lens calibration for ~15 meter detection

​

Pod Recovery: Recognition and Lock On

• IR Sensor + Range Finder
• While the Pixy Camera can detect the IR-Beacon, in order to have the drone lock on target, it needs a range finder using sonar or LIDAR
• Test in progress with Lidar Lite v.2

​

https://www.sparkfun.com/products/retired/13680

Pod Recovery: Precision Landing Mode

• Landing w/ Copter 3.4.6
• Copter 3.4.6 Platform version supports precision landing mode with set of parameters
• Flight Modes include: Stabilize, Loiter and Precision Landing
• Once the IR Sensor detects/locks on beacon, switch to precision landing mode
• Land on beacon

​

http://ardupilot.org/copter/docs/common-rangefinder-lidarlite.html

Breakdown: Sonar

• Raspberry Pi+2 30 khz Transducers
• Class D amplifier and buffer
• Improved mobility via Ad-hoc connections and battery pack installation
• Analysis with Matlab and Audacity

Sonar Analysis

Team Sonar

Signal Processing — Software

Currently using Matlab with Bellhop

Goals:

• Simulate SONAR data gathering
• Process SONAR to reconstruct ocean floor

​

Generating Test Audio

Click for sound

Reconstruction and 3D

• Synthetic aperture SONAR

Timeline

Completed

Round 1 Competition

1. Finish SONAR hardware and Data Collection
2. Finish Winch based Pod System
3. Finish Signal Processing
4. Get Drone to fly Autonomously
5. Attempt recovery of pod

Round 2 Competition

• Get Drone to fly Autonomously
• Recovery of pod by Drone
• Hybrid Gas Electric Power
• Multiple Drones and Pods

Thank You

What is the Shell Ocean Discovery XPRIZE?

https://www.youtube.com/watch?time_continue=103&v=Vhx7NCSEpe4

IMU Module

• Accelerometer, Gyro scope and pressure sensors to approximate position underwater
• Data recorded can be used in post processing to help determine location of each SAS sample

IMU Module

• One technique is have a our pod revolving and measure the changes in centripetal acceleration
• Self contained modules in epoxy that have been deployed to depths up to 100+ feet in the ocean

More Info: Pod

• 28 kHz Sound system for SONAR
• Variable Buoyancy Diving System
• IMU + GPS based position sensors
• WiFi + LoRa Data communication
• Sensor for chemical signature

Pod Movement

• Uncertain pod movement in the ocean:
• SAS literature for motion compensation [Callow 2003, Caporale and Petillot 2016, etc.]
• Use on board sensor information, such as IMU and depth sensors
• Localization and reconstruction from robotics [Fallon et.al. 2011, Newman 2006]
• Topological data analysis from swarm robotics and sensor networks [Dirafzoon et. al. 2014]

​

​

Data Processing

• Significant computing resources required for processing data
• Use cloud computing to run multiple servers in parallel
• Low cost and improves processing speed
• Use signal processing techniques, such as compressed sensing

​

Underwater Rocket Tests

Chemical Tracing Pod Hardware

Mechanical Design

Marine Vehicle Features

• Lightweight, CNC’d rigid delrin truss construction
• Frame designed for righting torque with coaxial COM and COB
• Mounted Kort Nozzle Design implemented to increase thrust
• Fluid dynamics simulations and iteration to verify minimal drag

​

Mechanical Design

Pressure Tolerant Waterproofing

• External waterproofed wire design verified at 3000 psi
• Buoyant Epoxy Mix ratios and process figured out
• Adaptable, multi-part epoxy mold release technique figured out

Electrical

• Features
• DMA-based PWM signaling from the RPi, provided by the PiGPIO library
• Access to a 10-DOF sensor via I²C
• SSH via WiFi (support for both infrastructure and ad-hoc modes)
• Serial getty in case WiFi fails
• Future work
• Integrate the chemical sensing module

Software

• Features
• Entire infrastructure in Python
• Scripts for calibrating and driving the motors
• Future work
• PID control using gyroscope and accelerometer feedback
• Integration of a chemical tracing module
• Design and implementation of a fielding algorithm

Map Teams