The first step in the implementation of a CMAC-PID was the modeling of the vehicle in Simulink. Dr. Cheng Chin (Chin et
al. 2006)create a symbol library for
modelling of ROVs. The library provided a predefined models that can be change
by a simple script in Matlab.
As second step in the implementation of the CMAC-PID is to
implement a normal PID controller. The PID component of the system was
implemented in Matlab. The acceleration wanted is calculated with a script that
divide the input velocity by 8 and if the velocity wanted is equal to the
platform speed the acceleration is equal to zero .A value limiter was placed at
the output of the system to avoid saturation of the actuators. The tuning
values for the PID components were estimated giving as result:
Kp=[16 14 24 34 34 14];
Kpi=[0.01;0;0;0.01;0.01;0.001];
Kdl=[0.002 0 0 0.001 0 0];
The implementation of the CMAC component was by the constant
call of a script at a sample time of 0.01 seconds. A CMAC was implemented by
each one of the DOF’s .In the first 100 cycles is calculated the maximum and
minimum data points. As second step the neural networks is started. For each
cycle the weight values are adjusted. The adjustments values for the CMAC are
m=5,
and
.Appendix
3 c). Figure 63 show the respond to a step input in all
DoF. The respond of the PID is calibrated to the fastest answer with a minor
overshoot. In rotation movement and vertical motion, the system show low overshooting
and fast stabilization. In lateral motions it is notice a overshooting. However
the overshooting can be solved reduce the acceleration rating of the robot. Figure 65 provide the different between CMAC-PID
and a normal PID in the principal signal with overshooting. The result show a
minor change on the signal and a reduction on the overshooting. If the step
signal is change for a sinusoidal signal the CMAC-PID show a higher importance
in the maintenance of the resonance frequency of the system over the time.