From Figures 5-8, we can see that with output-feedback controller, the state values are larger and the rate of convergence is faster than those with state-feedback controller.
(34 37), {β ProbV, β QrobV, β ProbA, β QrobA } > 0, are the gains of the controller, whose initial values are mentioned in the Appendix.
On the other hand, with the non-switching controller, all global minimum values are between 1.70×10−09 and 1.44×10−06 (the mean is 1.84×10−07, and the standard deviation is 2.70×10−07).
Therefore, the neural controller weights and bias values are obtained from the MATLAB environment as a starting point for hardware implementation.
For the synthesis of controller, the fault time, patterns and values are unknown in advance, as motivated from a practical spacecraft control application.
The learning algorithm uses a new self-organizing rule that has been specifically designed to learn a controller mapping even when succeeding sensor values are correlated.
The controller's proportional, integral and differential values are nonlinear functions of error function that the algorithm is simple and easy to realize.
The genes with a higher M11 values are most sensitive to (controller of) faster changes, whereas the genes with higher M1 values are responsible for slower changes.
*) As the number of elite controllers in this study is only 3, all three single values are depicted instead of the median.
