sUAS Design for Sensory and Swarm Missions
This project outlines my efforts to design a workhorse UAV for Northeastern UAV (NUAV). Historically, our student group has used a heavy-lift octocopter to perform autonomous missions like the location of rockets. However, testing autonomous missions with such a bulky airframe became cumbersome, especially when payloads and tests did not require such a powerful UAV. Moreover, the club was in dire need of a smaller, fully autonomous vehicle for testing stereo cameras and swarm missions. Therefore, out of inconvenience and the need for an economic but reliable airframe, the initiative to design a multipurpose developer drone was born!
Development began in late December of 2019 With the intention of testing power systems to accommodate between 8-10″ tri-blade propellers. The airframe would also have to accommodate a Pixhawk Mini for flight control, NVIDIA Jetson Nano for onboard computation, and an Intel Realsense D435 depth camera for obstacle avoidance/landing missions. Knowing our group’s use of this airframe would involve aerial deployment, I decided to make the design as flat as possible to allow for stacking or other space-efficient methods of payload storage aboard a larger UAV in the future. Furthermore, I wanted to make this frame a cost-effective but versatile learning tool for our club so early designs relied heavily on custom carbon and 3D printed components.
The first task was modeling the base plate of the frame. This parted needed to be dynamic to test different arm sizes for various motor/propeller configurations. It also had to accommodate mounting points for all electronics, motor arms, and through holes for battery/payload plugs under the frame. Lastly, this early design was driven by the need to support a variety of payloads, such as compatibility with both the Zed 2 and D435 stereo cameras, and therefore ended up with a spacious interior. This plate would also have to be compatible with a custom PCB for power distribution and signal routing that the electronics group of NUAV would develop.
The top plate of the drone was similarly designed with the intention of being water jet cut out of carbon fiber sheets. Aluminum standoffs were used to separate the plates to make room for a rear-mounted Jetson Nano flight computer and forward depth camera(s). Equations were used to drive the length and orientation of the motor arms base on the propeller radius (outlined in the sketch above). As would turn out to be a terrible idea, I initially thought it best to 3D print the arms in order to allow quick production of different sized models or configurations. This decision was guided by the thought that these vehicles would not have to endure crashes, but oh how naive we were.
I began the design of a fully parametric 3D printed motor arm which would take user inputs for propeller radius, cross-sectional width, and motor mounting hole dimensions. To maximize structural integrity, I designed bulky geometries with the intention of printing with thick walls but lower infill percentages to reduce weight. The orientation of the arms needed to be designed such that forward-facing cameras would have an unobstructed field of view, therefore making a ‘squashed-X’ configuration the chosen method (another fatal error with the V1 design that I will comment on later). The arms would also house the motor speed controllers (ESCs) and wiring in internal channels that would connect to the power distribution board inside the frame. Lastly, a simple interface with the frame was designed using heat-set inserts within the arms that would allow them to be sandwiched between the top and bottom plates. Initial prototypes were FDM printed using high strength ABS and polycarbonate.
With the arm assemblies complete, I designed the remaining miscellaneous fixtures for the flight computer, GPS, battery mounts, and RC communications to conclude the first prototype of the frame. Using a ProtoMAX waterjet cutter, the top and bottom plates were cut out of 2.5 mm carbon fiber plates.
At this point, the PCB order had arrived and wiring of the Pixhawk motors could begin. In order to allow for quick swaps of different motors in the field, MT60 connectors were used on the phase wires of the motors. Other features to support the first test included the installation of an FPV camera, bottom facing XT60 plugs to simplify the battery wiring, and use of ArduCopter firmware/waypoint missions. The assembly and maiden flight of the first prototype can be seen below!
The first hover test was rather lack-luster with bad oscillations in the pitch and roll axis. Likely due to the irregular shape of the propeller configuration, this poor performance on the stock tune was an initial indication of deeper issues with the design. Nevertheless, I was able to tune out most of the oscillations and improve yaw stability by adjusting PID and rate parameters in Mission Planner. This gave our team the ability to perform the first set of missions which included endurance tests to compare motor-propeller configurations and a simple RealSense data collect.
To determine ideal motor-propeller configurations, 4 tests were performed on 8-10″ and 13″ propellers. For each test, a single 6s 2200 mAh LiPo battery was used to perform a basic 4-waypoint autonomous mission. The vehicle was set to fly the circuit continuously until the battery was discharged below nominal voltage (22.2 V). Voltage degradation results were plotted with respect to time as shown:
As is to be expected, the 13″ propeller configuration using T-Motor 4012-480 Kv brushless motors outperformed the 8-10″ variants. On average, the 13″ configuration resisted voltage drop longer than the others, leading to longer flight times. It’s handling and lift capabilities were also superior with tuning which solidified the choice to use them on future builds. In comparison, the RealSense data collect tests produced far from desirable results due to prominent vibrations and issues with the rosbag used to export message data. It was therefore of interest to develop a damped camera mount for the RealSense to remove jello and other vibration-related artifacts from the feed.
Design began using plate-mounted grommets to isolate a camera holder from the frame. This was somewhat difficult as the height of the frame was barely enough to facilitate the amount of camera tilt that we needed. Here also, the mount would have an adjustable angle to compensate for the forward tilt of the quadcopter in flight. Lastly, the tilting mechanism would need to give access to USB-C plugs in the side of the stereo cameras. According to these criteria, I developed forward and downward facing camera mounts for obstacle avoidance and landing missions.
Each type of camera would have a custom 3D printed sled to allow it to interface to a standard vibration damping system. This system consisted of 4 pairs of offset carbon plates, separated by vibration damping grommets, which isolated the central camera pivot point from the frame’s plates. The central pivot points used 2 circumferential screws to adjust the camera’s angle while leaving a cavity to pass wires through for interfacing with the camera. Lastly, I developed a downward-facing mount with similar isolation to aid with autonomous landings. Flight tests with these damped mounts successfully removed vibration artifacts and allowed the following landing and flight data videos to be recorded:
At this point, the project had spanned from December 2019 – March of 2020 with the final tests being the RealSense data collects. Though successful in flight, tests showed that the V1 had number of issues that would make its mass production impractical. The first of these was its poor flight performance. As mentioned, the ‘squashed X’ motor configuration gave the drone poor flight characteristics and made it very difficult to manually control due to a lack of pitch authority. Second, my design of the parametric, 3D printed arms became unnecessarily complex and as inferred previously, did not perform well in crashes even when made out of PC. Moreover, printing and removal of supports took way too long which was the nail in the coffin that motivated me to only pursue full carbon arms moving forward. Next, the overall structure of the frame was too wide and complex. Because the propellers were not mounted in a square configuration, the base plate’s relations became too complex to reliably change. I had also allotted an unnecessary amount of room for sensors that we would never actually test and designed far too many screws to hold the top plate on. This made servicing the vehicle in the field very inconvenient. A better version was needed to simplify the frame’s geometry while leaving room for only the electronics that we would need for future missions. Lastly, there was no longer a need to support different motor types as the 13″ configuration was the favored choice moving forward. With that and the transition to virtual work at the end of the Spring 2020 semester, I began the design of the V2 swarm drone.
When I started to design the next version of the frame, I made a definitive shift to using multibody parts in SolidWorks. This allowed me to create frames in the same part file as mounting components and other fixtures which drastically simplified larger assemblies and made parametric design easier. The V2 frame would be an assembly of parts but with no dynamics, so multibody modeling would be perfect as the frame itself would essentially be a rigid body. Using this method, I was able to design all of the frame parts in 1 file with a single set of equations driving multiple bodies.
Regarding the actual design of the new frame, I decided on a ‘true X’ or symmetric, square motor configuration in order to give the best flight performance with simple geometry. The new base plate was less spacious to only mount mission-critical components and utilized fewer stand-offs to make field maintenance easier. Many improvements were also made to the stereo camera mount including a downward cutout to allow a single camera to look ahead and downwards as well as frame-mounted vibration grommets to decrease the height of the damped camera mount. Cutouts were also made in the top plate to allow a bottom-mounted GPS to protrude past its surface. This eliminated the need to attach components to the top plate that had fragile wires which were often damaged during maintenance of the V1 design. Lastly, the height of the frame was reduced by a further 5 mm in pursuit of a more deployable platform for use on swarm missions. The driving sketch for the V2 frame and its respective multibody part can be seen below
Another crucial improvement of the V2 over the V1 was the decision to use tubes of carbon fiber for the motor arms instead of 3D printed parts. Here, 1″ OD, 0.875″ ID square carbon tube were used for ideal structural integrity, weight reduction, and simplified mounting conditions for the frame and motor-mount interfaces. To fix them to the frame, I wanted to minimize the use of 3D printing to create only simple geometries that would share the forces during a crash with stronger carbon and aluminum components of the frame. To achieve this, I designed simple 1-piece clamps that bolted through the carbon arms and screwed between the top and bottom plates using heat-set inserts. These brackets were small, lightweight, easily printable, and tight fitting on the tubes to avoid any play in their assembly. In addition to the clamps being sandwiched between the frame, 2 adjacent standoffs were also placed tangent to each carbon arm to distribute force to the frame and away from 3D printed components during crashes.
The next step of the design was to create various fixtures such as motor and electronics mounts. Now that the shape of the frame was finalized, I began work with our project’s electronics group on a new PCB for the V2 airframe. The V1 PCB had many unused signal and power traces so simplification was needed. The PCB would mainly be for power distribution to the motors and key 5 V components while only routing crucial signal wires such as those for the ESCs. Simultaneously, I guided and finalized work by our hardware-design group to create motor mounts, a new Jetson Nano case, and articulating stereo camera mount for the RealSense. Pictures of these design steps are shown below:
The last component of the frame that I designed was the landing gear. In comparison to fragile, 3D printed designs of the V1 frame, I wanted these to be made out of sandwiched carbon plates to be strong with minimal added weight. I also made these components detachable to prepare for missions that wouldn’t require them such as the landings that we intended to pursue in researching deployable UAV swarms. In addition to extending the frame to the ground, this landing gear design used TPU printed pads to maintain traction during takeoffs from smooth surfaces.
With the completion of the frame and its components, the manufacturing of the first prototype began with waterjet cutting the frame plates and FDM printing parts.
We used laser-cut wood to test the plate dimensions and make a few minor revisions before waterjet cutting out of carbon. Tolerances were also tested and iterated with the motor and arm clamps. Overall, the assembly was largely ‘print-and play’ with only some experimentation needed with materials for the motor clamps before nylon prototypes could be manufactured. Wiring was also quite clean given the placement of regulators and use of the carbon tube arms to internalize motor and ESC wiring. The clamp system was exceptionally rigid and while the build was tight, space was efficiently used as per the design goals. The overall weight of the airframe was ~2.3 kg. This was about the same weight as the V1 but with a much more accessible, efficient, and aerodynamic design. Following a successful assembly, flight tests were conducted using a similar waypoint mission to the V1 efficiency tests to evaluate the performance of the new design.
With a stock ArduCopter tune, the vehicle was completely stable! Flight tests showed consistent tracking of waypoint missions with no oscillation and clean turning. The V2 design had successfully eliminated the poor handling characteristics of its parent designs while retaining all functionality. Furthermore, the landing gear grips worked to give added traction for more reliable manual/autonomous takeoffs. Crucially, endurance tests showed a capable flight time of 10-12 minutes which was up to a 3 minute improvement on the first design (within the same weight class).
With the functionality of the initial prototype verified, the beginning of the fall 2020 semester marked a manufacturing spree to produce another 5 vehicles. Over the course of 1 month, myself and the other co-founder of NUAV fixed design bugs, integrated electronics, and ultimately completed 6 vehicles in total to support autonomous tests.
Moving forward, this airframe will be an essential tool for NUAV members to learn software and hardware development skills in pursuit of our Swarm Carrier Project: to deploy swarms of UAVs from aerial platforms for rapid surveying missions. You can read more about this initiative in my post on Concept Drone Deployment from an Aerial Platform.