Aerospace and Electronic Systems Magazine June 2017 - 63

Babb and Yilmaz

Figure 3.

The fuzzy membership functions for the (a) x-axis error, (b) x-axis error
rate, throttle changes for the (c) right motor, and (d) left motor (the yaxis ranges 0-100%, while the x-axis ranges −180 to 180 deg).

troller input x-axis error and error difference variables were fuzzified by using five and three membership functions, respectively,
as shown in Figures 3a, 3b, to reflect the practical operation conditions and steady-state accuracy. Similar membership functions
were used for the y-axis error and error difference inputs. Because
the MCU library supported specific membership functions, such
as triangle or trapezoidal types, the same types were used during
the controller synthesis. The FL controller outputs, i.e., the throttle
changes for four motors, were fuzzified by using seven membership functions with different types, as shown in Figures 3c, 3d,
assuming similar designs for the motor pairs. The remaining two
outputs along the y-axis included similar membership functions.
Note that the higher number of membership functions offers resolution in the fuzzy variable but also complicates the inference rule
development efforts. The FL controller inference rules were designed based on human experiences of quadrotor system operations. The higher number of fuzzy inference rules may increase the
overall performance of the FL controller. However, determination
of accurate and relevant inference rules may also complicate FL
controller synthesis.

EXPERIMENTAL RESULTS
The C- and assembly-based programming challenges, FL controller synthesis, and electrical and mechanical implementations indicated the need for extensive background, as well as debugging
skills with the HCS12 microprocessor and an interdisciplinary
team for a successful experimental implementation. The feedback
control system performance was first tested by using the simplified quadrotor dynamics, with expected iterations. The controller
software was implemented by developing the corresponding code
in the assembly language, downloading the software on the MC9S12DG256 microcontroller, and debugging the associated programs
for satisfactory quadrotor operations.
A quadrotor and test rig were built. The quadrotor was affixed
to a pedestal-style test rig and was configured to allow rotation on
only the pitch axis of the quadrotor. The test rig allowed tilting
within [−15, 15] deg of horizontal and held the quadrotor off the
ground by about 4 ft.
The AirDragon real-time operating system software was improved to include both a flight data recorder and a serial comJUNE 2017

Figure 4.

Data capture of (a) the platform angle relative to the horizon, (b) the
FLC error correction output, and (c) the FLC error rate correction output (the y-axis values are listed, while the x-axis ranges 0-4 seconds).

munications service (SCS) routine with the newly written serial
communications interface driver. This enhancement allowed the
MCU to transfer data to the Xbee wireless communication modules, permitting one-way data transfers from the MCU directly to a
personal computer (PC) running HyperTerminal.
Two successful trials were performed, one for "instantaneous" data capture of roughly 4 seconds of data and one for
"continuous" data capture of roughly 60 seconds of data. The
FDR was configured to capture the angle data and the output of
the error and error rate Fuzzy Logic Controller (FLC)'s, and the
SCS was configured to dump the capture data over the Xbee to
the PC after a test run had been halted. For both experimental trials, the quadrotor pilot program was configured to run a simulation lasting approximately 60 seconds, if the kill switch was not
pressed during the trial. If the kill switch was pressed or disconnected during a run, the quadrotor would immediately stop all
motors and dump whatever data it had collected over the Xbee.
The instantaneous data capture had also configured the FDR to
capture the data of each cycle of the FLC algorithm. Due to limitations, the FDR could only hold up to 600 lines of three-column
single-precision floating-point tables. Therefore, the instantaneous data capture could only capture roughly 4 seconds of data
for a 110-Hz FLC operating frequency. The instantaneous experiment was performed several times, and the corresponding sample
system responses are shown in Figure 4 for the experimental
quadrotor-captured angle, error FLC output, and error rate FLC
outputs. The FLC's preprocessing block saturates the calculated
error to be within ±10 so that it can be mapped to an 8-bit format
that is used in the current FLC implementation.
The output PWM signals to the ESC on the quadrotor had
to be reconfigured to be at least 2.2 times faster than the control
loop to ensure that the output of the error rate controller would

IEEE A&E SYSTEMS MAGAZINE

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine June 2017