Robust data-driven feedback linearization using neural network based sparse identification of nonlinear dynamics

Achieving robust data-driven and high-precision control for nonlinear systems is a critical challenge, particularly in the presence of model uncertainties,unmodeled dynamics and disturbances. In this work, a data-driven control framework is proposed by integrating Sparse Identification of Nonlinear Dynamics (SINDy) with feedback linearization. An accurate and low-dimensional representation of the system dynamics is extracted directly from data using SINDy which reduces reliance on explicit first-principles modeling. The identified model is then utilized for state feedback linearization, enabling precise control and robust tracking. simulations have been conducted on benchmark nonlinear systems, including an Inverted Pendulum and a Single-Link Flexible-Joint Robot Manipulator. A comparative analysis with state-of-the-art controllers demonstrates superior tracking accuracy, faster convergence, and enhanced robustness with minimal computational overhead. Despite the effectiveness of the proposed approach, residual modeling errors, uncertainty and external disturbances remain present. To address these challenges, a neural network-based observer is introduced, allowing for the dynamic estimation and compensation of unmodeled dynamics. This enhancement significantly improves stability under uncertain conditions. The findings confirm that data-driven feedback linearization offers a scalable and real-time applicable solution for complex nonlinear systems. By bridging model-based and data-driven control, this framework establishes a foundation for next-generation data-driven robust control strategies.