Helicopter controlling and balancing

The Cerebellar Model Articulation Controller (CMAC) is a neural network inspired by the neurophysiologic theory of the cerebellum. The CMAC was rst described by Albus [1, 3] in 1975 and despite its biological relevance, the main reason for using the CMAC is that it operates very fast, which makes it suitable for real-time adaptive control. According to the control scheme proposed by Miller [15, 16], the CMAC learns the inverse dynamics of the plant while it is controlled by a classical controller. This makes the training of the CMAC memory unpredictable because for a particular control setting, the plant output typically follows a certain trajectory. Thus, which particular memory cells will be covered by the plant output trajectory is undetermined. Therefore, the learning phase of the CMAC has to be planned carefully to ensure the entire characteristic surface is trained. In addition, since the number of memory cells in the CMAC is finite, the control output is discrete, which results in heavy fluctuations in the system. Increasing the memory can be a solution to this problem but it is not always feasible. As a result, the Modi ed Cerebellar Model Articulation Controller (MCMAC) was proposed in [17] to overcome these limitations. It successfully removes the conventional controller and at the same time, achieves very good performance [4]. Moreover, the Averaged Trapezoidal Output (ATO) was also proposed in [4] and incorporated into the MCMAC to reduce the e ect of the quantization error without using extra memory cells. Therefore, the MCMAC is designed and developed to control the pitch axis of a real 2 DOF helicopter built by Quanser. The results obtained from many experiments show that its performance exceeds those of the CMAC and the supplied LQR. It is on a par with the Sliding Mode Control (SMC) via LQR and sometimes, it is even better under certain conditions.Bachelor of Engineering (Computer Engineering