|
3. CONTROL PERFORMANCE VERIFICATION BY SIMULATION
3.1 Expansion of the existing simulation technique to two floating bodies
A dynamic positioning analysis and evaluation program for two bodies has been developed based on the dynamic positioning system analysis program DP-MAP[2],[3] suitable for single floating body. The following improvements were undertaken in order to extend this program
(a) Development of a mathematical model for the maneuvering motion of two bodies
(b) Development of a mathematical model for the hawser that connects the two bodies
(c) Development of neural network control and learning logic.
The extended flow of the program is shown in Fig.3. The dynamic action model of each body follows the conventional model. In addition, the hawser tension is obtained semi-statically from the characteristics between tension and hawser length.
Fig.3 |
Flow chart of DP-MAP dynamic positioning program for coupled floating bodies |
3.2 Results of simulation
The control system was included in the above-noted dynamic model for two bodies, and simulation calculation for the performance evaluation of this neural control was carried out. The basic calculation items under the external forces shown in Table 1 are as follows:
(a) No control, (b) PID control, (c) Neuro control
Table 1 External force conditions for relative position control of coupled bodies
| Irregular waves |
H1/3=4.5m
Tp=6.5sec
JONSWAP Spectrum
(peak enhancement factor γ=3.3)
β=0deg |
| Wind |
none |
| Current |
Uc=2.0m/s
ψc=15, -l5deg |
|
The results of computation without control conditions and PID control are shown in Figs. 4 and 5 respectively. The PID control gains were decided as if only an FPSO was operating under yaw direction control. Even if the FPSO is performing sufficient direction control, hawser tension became excessive due to the coupling motion with the shuttle tanker, and the control limit was encountered in direction angle PID control of FPSO. That is, for a strong disturbance of 4.5m in significant wave height, there are few differences between PID control and non-control. Effective results were not obtained in terms of reducing the number of times that hawser tension exceeded 100 tons.
As the peak value of hawser tension decreases in comparison with PID control shown in Figs. 6 and 7, this neural network control method can be said to be effective. As the neural network learning data was simultaneously obtained from dynamic motion, hawser tension peak values were effectively controlled to within 100 tons after 1500 seconds when sufficient learning data was obtained for the control system.
Thus, it was shown by simulation study that neural network control can lower the hawser tension.
Fig.4 Simulation results without control
Fig.5 PID simulation results
Fig.6 |
Neural network control simulation results with continuous learning |
Fig.7 |
Neural network control simulation results with pre-learning |
Fig.8 Coordinate system in the experiment
|
