Abstract:
This thesis presents a neural network-based framework for modelling, estimation, and
control of a DC motor system with hard nonlinearities. A detailed motor model
incorporating damping, Coulomb friction, and backlash is developed, with parameters
estimated using systematic methods validated by prior research. The motor is then
modelled using Time-Delay Neural Networks (TDNN) and Nonlinear AutoRegressive
models with Exogenous inputs (NARX) to create network-based observers, which are
compared against classical observers for estimation nonlinear dynamics. A novel structural
enhancement is introduced into the TDNN, addressing challenges such as error
accumulation and nonlinearity-induced discrepancies, achieving superior accuracy and
robustness over traditional approaches. The structured TDNN is integrated into a control
framework and compared with a classical controller under dynamic conditions,
demonstrating better adaptability, faster response times, and improved tracking accuracy,
particularly in scenarios dominated by backlash effects. This work contributes to neural
network-based control by providing a robust methodology for handling nonlinear systems,
highlighting the potential of AI-driven solutions to replace conventional control
approaches in motor systems and beyond.
