An AI‐Enhanced Adaptive Discrete Sliding Mode Control Framework for Non‐Affine Nonlinear Systems

ABSTRACT Despite many years of discrete sliding mode controller (DSMC) development, non‐affine nonlinear systems remain fundamentally intractable due to nonlinear control coupling that violates the DSMC's affine structure requirement. This limitation excludes vast classes of real‐world systems, including chemical reactors, aerospace vehicles, and biological processes, from benefiting from DSMC's proven robustness. This work successfully addresses the long‐standing challenge of applying DSMC to non‐affine systems, motivated by the critical need to extend robust discrete control methodology to nonlinear non‐affine systems. Unlike previous NN‐based DSMC models, our method provides global convergence and robustness without requiring offline training or fixed‐gain design. This work introduces three novel contributions: (1) first‐ever integration of least‐squares support vector regression (LSSVR) with DSMC for non‐affine systems, (2) novel NARMA‐L2 decomposition enabling exact affine transformation for DSMC, and (3) dual‐layer adaptive framework with proven global convergence. First, the nonlinear autoregressive model with exogenous inputs (NARX) framework is employed, with online LSSVR applied to identify system dynamics accurately. This NARX model is then transformed into a NARMA‐L2 structure, which allows the original non‐affine system representation to be reformulated as an affine system. Using this affine transformation, a discrete‐time sliding mode control law is derived and formulated based on system dynamics to ensure robust trajectory tracking under parametric uncertainties, external disturbances, and measurement noise. The approach includes adaptive tuning of DSMC parameters using the Levenberg–Marquardt algorithm, allowing real‐time updates to controller gains based on observed errors and model discrepancies. Simulation results demonstrate that the proposed control architecture provides superior tracking performance and robustness across a range of challenging conditions, showcasing its potential for application in high‐precision, digital control systems. This integration of adaptive DMSC and LSSVR within a NARMA‐L2 framework represents a significant advancement in sliding mode control for non‐affine systems in discrete time.