Black-Box Identification and Control for Autonomous Underwater Vehicles

Abstract

Modelling and control of Autonomous Underwater Vehicles (AUVs) pose serious challenges due to their complex, inherently nonlinear and time-varying dynamics. Thus there is a necessity for designing a robust and stable control system that would selftune the system when the performance degrades during the operation. Due to the complexities associated with the AUV dynamics, a black box identification technique is used for modelling the behaviour of the AUV. To improve online identification in the presence of noisy data, this thesis introduces a novel identifier scheme for identification of non-linear systems with disturbances, that we call semi serial-parallel model (SSPM). The model based controller and the identification model of the UNSW@ADFA AUV are developed using the offline and online techniques and are based upon Fuzzy system and Hybrid Neuro-Fuzzy Network techniques. A novel auto-generating mechanism with entropy based Differential Evolution (DE) system modelling is proposed to generate the AUV model without any prior knowledge of the physical relationship inside the system or its behaviour. It is carried out in two steps; an off-line procedure and an on-line procedure. The method comprises of an automatic structure generating phase using entropy based technique and a parameter-learning phase. The accuracy of the model is suitably controlled using the entropy measure. The parameter learning phase uses the back-propagation technique. To improve the accuracy and also for generalization of the model to handle different data sets, DE technique is employed whereby the parameters of the model are suitably tuned using evolutionary technique. A number of benchmark problems are considered to validate the methodology and the accuracy of the results from the proposed method are found to be superior compared with the existing techniques. The proposed mechanism is used to design the control system. These techniques need to be real time implementable. Real time validations using Hardware In Loop (HIL) simulations show that both identification methods are feasible for control. The accuracy of identification using online auto-generating mechanism is better than the offline technique in terms of accuracy and computational time. HIL simulations applied to simulate the performance of the developed control system of surge, pitch and yaw movements. Finally, real time experiments of the proposed algorithm for identifying the AUV dynamics are implemented. The experimental results show that the proposed identification mechanism and the model adaptive controller are capable of controlling the AUV suitably in a real environment, demonstrating its robustness characteristics.