This paper proposes a hybrid intelligent control strategy for high-performance induction motor drives under field-oriented control (FOC). The proposed architecture combines a radial basis function (RBF) neural network with higher-order non-singular terminal sliding-mode (HONTSM) control to improve robustness against nonlinear dynamics, parameter uncertainties, and external disturbances. The RBF network generates the torque-producing current reference in the speed loop, while the HONTSM compensator enhances disturbance rejection and tracking performance. In the inner loop, a HONTSM current controller ensures fast convergence, reduced chattering, and robust current regulation. To further enhance energy efficiency, an improved Kronecker-Factored Approximate Curvature (IK-FAC) flux optimization strategy is proposed to adaptively optimize the flux-producing current reference in real time. The proposed approach employs recursive curvature estimation based on filtered gradient-energy statistics to adaptively scale the flux update law without explicit Hessian computation or matrix inversion. Unlike the original K-FAC algorithm for large-scale neural network optimization, the proposed method reformulates the curvature-aware adaptation principle into a lightweight scalar recursive framework suitable for real-time embedded motor-drive applications. Consequently, the proposed IK-FAC strategy achieves fast convergence toward the minimum-loss operating trajectory while preserving the original FOC structure and maintaining low computational complexity for DSP-based induction motor drives. Controller parameters are optimized using particle swarm optimization (PSO) based on the integral absolute error (IAE) criterion. Experimental validation on a Texas Instruments TMS320F28379D digital signal processor demonstrates improved tracking accuracy, fast transient response, enhanced energy efficiency, and strong robustness under varying operating conditions.
Ngoc Thuy Pham (Fri,) studied this question.