Final Design

Final Design Solution

    The final design solution is a single rotor duct drone made out of carbon fiber that is capable of maneuvering using a thrust vectoring system to redirect airflow, allowing the system to pitch, yaw, and roll. A CAD model of the design can be seen above. The main components of the design consist of the carbon fiber frame, the counter-rotating propeller system, and the thrust vectoring control system. The carbon fiber duct with a uniquely calibrated curvature on the top allows for additional lift to be generated by the Coanda Effect. The Coanda Effect describes the effect in which high-velocity air moves over the top of the duct creating a low-pressure area, which is met by high pressure underneath the lip. This high-pressure zone generates additional lift for the system. The counter-rotating propeller system is what allows the drone to generate approximately 10kg of thrust while restricting unwanted yawing motion. Having two propellers rotating in opposite directions creates equal momentum in opposites direction, thus preventing the duct from rotating about its center axis. Finally, the thrust vectoring system allows the drone to be maneuverable by redirecting airflow, allowing for control of the pitch, yaw, and roll of the drone. The thrust vectoring system is made up of a combination of 4 rudders, each independently

Design of Key Components

Carbon Fiber Duct

Functional Requirements

The purpose of the duct in Project Laputa is to increase the efficiency of the motor and propeller systems, and to generate additional lift. A duct increases the efficiency of the motor and propellers by stopping any wingtip vortices that may be generated at the edge of the propellers. Wingtip vortices are comprised of turbulent, rotating air that trail off the edge of lift-generating surfaces, and causes induced drag. It was necessary to fabricate the duct under tight tolerances in order to have a minimal amount of space between the edge of the propellers and the duct without contact, so that wingtip vortices are not able to form. By curling the upper edge of duct outwards, additional lift can be generated via the Coandă Effect. Additional air is funneled towards to propellers, and the top and bottom of the lip itself experiences a pressure difference due to the faster flowing air above the lip. These two effects bring additional lift to the drone system. The shape of the lip determines how much additional lift can be generated. Finally, the duct itself needs to be strong as it is the main frame that all other components will be mounted too. However, the duct also needs to be as lightweight as possible in order to maintain the best weight-to-lift ratio for the whole system.

Comparison of Designs Considered

    The possible designs for the duct only vary in terms of lip shape. All other design considerations to meet the functional requirements stay the same throughout all designs. Those designs considerations include a tight tolerance between the inner duct and the propeller, as well as limitations on possible materials. Carbon fiber is the best possible material, followed by fiberglass. Other materials do not offer the same strength while maintaining a low weight. All lip designs considered are listed under Appendix E in the final report.

Justification of the Final Design Choice

Figure 1: Final duct design

    The final curvature was decided on by running Ansys simulations on 13 different 3D modeled designs of the duct, which were similar variations to those seen above. In addition to this, research was done on similar ducted fan designs to determine what has worked in the past. In most cases, many designs were eliminated because they did not produce enough lift. However there were a few cases in which the design was ruled out due to difficulty of manufacturing (the 270° circular curvature is an example of this). The modified lemniscate curve design produced the most lift according to the ANSYS analysis as well as previous reports. Furthermore, the height of the duct has no effect on the functionality of the duct, and only serves to increase weight if increased.

Figure 2: Cross section final duct design

The final design consisted of a properly scaled lemniscate curve from t=[0.5, 1.5]parametrically, and a quarter circle arc. The straight length of the duct is determined based on the minimum amount of space all components (two motors, two propellers, base assembly structures, etc.) can be mounted in. Carbon fiber ended up being the material of choice in this application due to its high strength to weight ratio and its ability to be easily molded into any custom shape. Because of the advanced carbon fiber manufacturing technology that is available, it is possible to achieve very tight tolerances wherever they are needed. This is ideal for creating a highly precise and reliable frame.

Figure 3: ANSYS CFD analysis of the final duct design

The duct design used in the first prototype was analyzed through ANSYS Fluent, a computational fluid dynamics (CFD) program as a rough estimate of what airflow would be like through the duct. This analysis uses several assumptions and generalizations to determine boundary conditions for the sake of simplicity, such as uniform airflow. A low-velocity area (light blue to dark blue) can be observed underneath the lip in Figure 3, in the shape of an eddy. This represents the slow air that is trapped underneath the lip that continuously cycles in the same area, creating a comparative high-pressure zone. The high-velocity airflow (orange to red) can be observed at the upper side of the lip. This represents the faster-moving air that created a comparative low-pressure zone. Through this CFD analysis, the additional lift theorized to be generated by the Coandă Effect can be seen.

Fabrication of the duct was originally planned to be done through autoclaving prepreg carbon fiber that was laid on to a mold made out of high-temperature epoxy board. Due to financial restrictions, the duct for the current prototype was instead made using the wet layup process for carbon fiber. The mold was made out of CNC-capable foam boards that provide a smooth surface upon machining. The straight section of the duct was created using a large industrial lathe, and the lip section of the duct was carved using a CNC Routing machine. The two mold parts were joined using simple wooden dowels, and was then covered in a layer of teflon tape to make it as smooth and seamless as possible. Following this, the carbon fiber was hand-molded and let to cure at room temperature for several days.

Motor and Propeller System

Functional Requirements

    To ensure the ability to fly, a system must be able to provide enough thrust to lift the entire system and its payload. Providing enough thrust is key to the generation of lift for flight. The Laputa team aimed at developing a motor and propeller combination that is able to lift twice the weight of the entire system. The heaviest items of the system include the motors, Li-Po batteries and the basestation payload. The estimated mass of the entire system is roughly 5-6kg. Therefore the motor propeller system needs to provide enough lift to support 10-12kg. The key is to find the most efficient and highest quality combination of propellers and motors available. To ensure the thrust capabilities of the system and to counteract the rotational torque on the system from a single propeller design, a dual motor/propeller system will be implemented. The orientation of a dual propeller system must be investigated and designed.

Comparison of Designs Considered 

Considered Motor and Propeller Orientations

Table 2: Thrust propulsion system design comparisons

Justification of the Final Design Choice

    The dual motor and propeller stacked orientation was the final design choice for several reasons. Although the back to back coaxial system required only one mounting plate, the system still required a second platform for the purpose of mounting the thrust vectoring system. Additionally, the back to back coaxial system requires a “push” style propeller, as a result of the upside down mounted motor. Unfortunately, the company that produces the highest quality propellers on the market sold the push style separately from the pull style, which would have cost an extra $387, which was not in the budget. Plus the propellers are sold in counter-rotating pairs, which would leave the team with two extra propellers not in use. Therefore to remain at a reasonable price level and provide a mounting platform for the thrust vectoring system, the stacked coaxial method was chosen.

    The Laputa Team aimed at producing the most efficient, long flight oriented EDF ever made. To do this, the most efficient propeller and motor system is necessary. As the motor and propeller orientation was finalized, the actual components could be analyzed. With such a wide variety of motors and propellers on the market, there are many aspects to consider. Motors need to be fairly large and sturdy to support the weight and thrust of the system. A shorter, wider, more pancake-like motor will be better suited for our needs as it has a lower center of mass and will require less mounting space. Which in turn will decrease the overall height of the duct. Regarding power, all DC brushless motors’ power is defined by their KV rating. KV is not a kilovolt (kV) rating, it defines the rpm constant of the motor, i.e. the revolutions per volt. High KV motor have few winds of thick wire, spin very fast and draw a great deal of power. Low KV motors have more winds of thinner wire and therefore will carry more volts at fewer amps, produce higher torque, swing a bigger propeller and is much more energy efficient.

As for propellers there are several areas to consider as well such as the number, size, pitch and material of the blades. For the purposes of the Laputa project the lightest most energy efficient option would be large diameter, low pitch propellers made of carbon fiber. Then the size of the motor determines the size of the battery and the propeller determines the size of the duct. All these components factor into the weight calculation. Therefore, we will need to make weight assumptions to figure out the system requirements, which then will be used for the correct weight calculations and the entire optimization process.

It was ultimately decided to acquire the highest quality motors and propellers from a Chinese company known for their UAV systems, T-Motor. Selecting the combination of the MN805-S KV120 motor and a 30” (76.2 cm) carbon fiber propeller from T-Motor seemed ideal for the project’s requirements.

Figure 4: Motor support structure model and FEA analysis

Each motor is attached to the duct through four aluminum tubes and a solid aluminum plate. Aluminum tubes were used instead of rods to save weight. The aluminum plate is machined out of a half inch thick, five-inch diameter circular disc. Excess aluminum on the disc not needed for structural support for either the tubing or the motor is trimmed off to save weight as well, resulting in the shape seen. Detailed drawings for the motor support strut tubing and the motor support plate is located in Appendix C.

Control System

Functional Requirements

Project Laputa has many requirements. One of the main challenges of a dual rotor duct system is keeping it stable during flight. Ducted fan vehicles have complex aerodynamics which make them highly unstable. Therefore, for a controller to be successful, it needs to be very robust to account for the uncertain conditions it could be presented with. Additionally, the flight controller should have limits to the flight envelope (i.e. combinations of speed, altitude, attack angle, etc. within while flying stable).  The team has considered two designs for the flight controller hardware.

Comparison of Designs Considered

Table 3: Control system design comparisons 

Justification of the Final Design Choice

A thrust vectoring control vein system was chosen as the main stability and motion system for its preciseness and weight. When compared to the other considered designs, the thrust vectoring system is much lighter if utilizing thin 7075 aluminum or carbon fiber. No need to add extra weight of the metal sliders or winches, as these flaps can be control precisely with small and light servo motors. The trade-off is that this complicates the system firmware. Seemily impossible for MAE students to design and code such a system within a scope of a quarter, yet with the addition of an ECE team this suddenly seems viable. The current design remains with four independent flaps that are firmly attached to a rod that stretches from the outside of the duct to just under the mounting plate of the second lower motor. This allows the servos to be directly connected to the rods from the outside of the duct in order to not disturb the fluid dynamics in play. The control veins allow for complete control of roll, pitch and yaw of the system with the integration and tuning of a PID control system. The large thrust vectoring single rotor ducted drone aimed at long flight time has never been accomplished before and thus fulfils one of the sponsors requirements of novelty. The control veins and rods will be machined out of 7075 aluminum stock and sheets. The veins can be precisely cut from carbon fiber blocks if those will be available to the team. Still in the design process to see what material would work best and is most available.

Filtering Data

Functional Requirements

The raw data from a gyroscope and an accelerometer are in units of degrees/s and m/s^s. These measurements must be converted into angles for pitch, roll and yaw. Through integration and trigonometry, one can convert both the gyroscope and accelerometer measurements to angles in degrees. The problem after this step comes with correcting the data through filters. Gyroscope data tends to drift to different values over longer periods of time. Accelerometer data is prone to poor instantaneous values. In summary, in the long run, gyroscope data works better in the short run and accelerometer data works better in the long run. An ideal filter would utilize both these properties.

Comparison of Designs Considered

Table #: Filter Comparison

Justification of the Final Design Choice

Using the complementary filter on the sensor readings resulted in accurate and stable readings. Justification for the usage of a filter can be seen in the graphs below. The sensor was laid flat and kept at a constant angle. Two tests were conducted, one with a complementary filter and one without a filter. The drifting of the measurements without a filter derives from a concept called “gyro drift”. This is when the gyroscope measurement will start to drift away from its current value and become inaccurate after long periods of time. It results from a small error that builds up over time.

Figure 5: Pitch Angle with and without Filter

Figure 6: Pitch angle with Filter

From the first graph, the angle results between using a filter and not using a filter are much different. The angle without a filter drifts to more than -15 degrees after 3.5 minutes while the angle with a filter does not seem to have any significant drift. In the second graph, one can observe how there is slight drift in the angle values to almost .5 degrees after 3.5 minutes. This drift is very small in comparison. It is almost impossible to completely eliminate gyro drift from a system, but continued tests of different filter values will result in an angle value that can be trusted.

Control Algorithm

Functional Requirements

A control algorithm needs to quickly detect and correct the position of the drone. The control algorithm must correct pitch, roll and yaw in order for the drone to maintain its stability. Two main tradeoffs in a stability system are oscillation versus rise time. The algorithm would ideally eliminate oscillation and still maintain a small rise time that can correct angles. Many hobbists and engineers use PID control algorithms.

Comparison of Designs Considered

There are several control systems similar to PID. Two commonly used options are PI control and PD control. Both systems act similarly to PID control, except they have less components and are therefore simpler to implement. Considerations for performance are oscillations, rise time and ensuring that there is no error between the steady state position and the offset position.

Table #: Control Comparisons

Source: http://www.ni.com/white-paper/6440/en/

Justification of the Final Design Choice

    Based on the comparisons above, PID has the best performance. The PI control would not be sufficient for this application because the response times for these types of controllers tends to be too slow. The drone would not be able to react fast enough to sudden changes in its environment. The PD control would not be sufficient for this application because this type of controller would tend to have offset error when it reaches a steady state. This means that the desired angle of this drone would be slightly tilted and therefore would be moving away from its desired location. This drone is intended to be airborne for as long as possible (> 1 hour) so this slight offset will move the drone significantly over this quantity of time.

The PID uses proportional, integral and derivative components to correct an error. A detailed description about how PID works can be found in the Appendix.

Testing PID Control:

1) P Control Only:

Figure 7: Testing PID Control:

    To test the PID control algorithm, the prototype was clamped using a stand on both sides of the rod. This way, external forces and disturbances can be introduced (i.e. physically pushing the duct) and the results can be observed. Although there is still some oscillation, the PID control proved to be successful. Upon physically pushing the duct, the algorithm knew it needed to write a specific speed to the servo motor. As a result of the servo motor changing speed, the flaps rotated in such a direction that pushed the duct back to the stable reference frame. This was all done by tuning Kp only. Ki and Kd were set to zero during this test. The reason behind this is so that Kp can be tuned all the way to the point where the system just begins to oscillate. Then, Kd will be tuned to eliminate the oscillation. This is because Kd affects the damping factor which affects the oscillation. Finally, Ki will be tuned to remove the steady-state error that is left. Please refer to Appendix H for details about the PID algorithm.