Rath, Biranchi Narayan (2020) *Robust and Adaptive Control Algorithms for Autonomous Underwater Vehicle.* PhD thesis.

PDF (Restricted upto 10/02/2023 ) Restricted to Repository staff only 4Mb |

## Abstract

The vastness of the ocean on earth provides limitless potential for exploration. This has sparked an interest in researchers to design and control underwater vehicles as a means to explore and use marine life more productively. Autonomous Underwater Vehicles (AUVs) nd a lot of applications including defence, pipeline survey and mine survey. However, the AUV dynamics is uncertain due to parametric variation, ocean current and wave disturbances. Hence, in order to address the motion control problem of an AUV, it is necessary to design robust and adaptive controllers. These motion control algorithms necessitate mathematical representation of AUV which comprises of hydrodynamic damping, Coriolis terms, mass and inertia terms etc. To obtain dynamics of an AUV, a system identication technique using Extreme Learning Machine (ELM) structure is considered in this work for the identifying the dynamics of AUV.

This thesis focuses on designing four robust and adaptive control algorithms namely, H1 model predictive controller (MPC), Fast Quadratic MPC (FQMPC), nonlinear model predictive controller (NMPC) and explicit model predictive control (EMPC) for accomplishing ecient path following control of an AUV. In order to verify the ef-cacies of the proposed control algorithms, both simulation and experimentation were pursued.

A prototype torpedo-shaped mono-hull AUV consisting of a single board computer along with Arduino microcontroller and with a thruster placed at rear end is designed in the laboratory. The proposed robust and adaptive control algorithms are implemented using the computational unit of the prototype AUV to achieve desired path following control of an AUV using python programming language.

The thesis rst focuses on the identication of the unknown dynamics of the AUV is identied by using an ridge regression based ELM model using the experimental input data (i.e. rudder orientation) and output data (i.e. yaw rate and sway velocity) from a prototype AUV designed in the laboratory. In order to increase the converge rate of the ELM model during training, the hidden layer parameters are optimized using modied Particle Swarm Optimization (PSO) algorithm. The problem of path following can be approached by parameterizing the desired path using Serret{Frenet(SF) frame. Using the desired path and kinematic equations of AUV, a Lyapunov based backstepping controller is designed to obtain the velocity references for the dynamics of AUV. This is subsequently used to design a dynamic H1 MPC to track the above velocity references. This H1 MPC algorithm is then simulated in MATLAB followed by real-time experimentation on a prototype AUV developed in our laboratory.
The uncertainties may arise due to wave disturbances, model mismatch, ocean current or payload change. A Fast Quadratic Model Predictive Controller (FQMPC) based on on-line sequential ELM (OS-ELM) is proposed to handle these uncertainties. To ease the control law formulation, way points tracking is used for controlling AUV for tracking a desired path. Thus, the FQMPC is proposed for addressing the path following control problem using way-points for an AUV.
The above two controllers namely, H1 MPC and FQMPC are designed based on linear dynamic model of AUV. However, the dynamics of AUV is nonlinear, so a nonlinear model predictive controller (NMPC) is then designed using OS-ELM model of AUV. In order to increase the ecacy in the performance of the OS-ELM model, the signicant terms in the input vector of OS-ELM model are determined using Forward Regression Orthogonal Least Square (FROLS) algorithm based on Error Reduction Ration (ERR) criterion. From the simulation and experiment results, it is observed that NMPC though provides successful tracking of the desired path, it is computationally expensive due to solving the optimization problem at each sampling instant.
Subsequently, to to reduce the computational time an Explicit Model Predictive Controller (EMPC) scheme is proposed. The EMPC is designed using two steps.
Firstly, using Generalized Hermite Biehler Theorem (GHBT) all stabilizing gains of a proportional-integral (PI) controller is computed o-line with a specied gain margin (GM) and phase margin (PM) design criterion. Secondly, based on the stabilizing set of PI controller, an Explicit MPC (EMPC) is designed using look-up table. The eectiveness of the proposed EMPC is veried by conducting experiment on the prototype AUV. The experimentation is conducted at a swimming pool with dimension of 10mx5mx1.6m. From the obtained results and evaluation of performances of all the proposed controllers, it is contemplated that the EMPC based control strategy is preferable for real-time implementation to steer the AUV to achieve eective tracking performances. [brace not closed]

Item Type: | Thesis (PhD) |
---|---|

Uncontrolled Keywords: | Autonomous underwater vehicle;machine learning;serret- frenet frame;way-point;model predictive control |

Subjects: | Engineering and Technology > Electrical Engineering Engineering and Technology > Electrical Engineering > Power Electronics |

Divisions: | Engineering and Technology > Department of Electrical Engineering |

ID Code: | 10146 |

Deposited By: | IR Staff BPCL |

Deposited On: | 10 Feb 2021 17:33 |

Last Modified: | 10 Feb 2021 17:33 |

Supervisor(s): | Subudhi, Bidyadhar |

Repository Staff Only: item control page