Model-Free Sliding Mode Iterative Learning Control for Cz Silicon Single Crystal Diameter

Abstract

Aiming at the complex nonlinear dynamic time-varying characteristics for Czochralski (Cz) silicon single crystal growth process and the difficulty in modeling and controlling the crystal diameter by conventional mechanisms, based on the idea of data-driven modeling and control, this paper proposes an improved model-free sliding mode iterative learning control (MFA-SMILC) method. First, a data-driven model of crystal diameter is established using an extreme learning machine (ELM) with actual process data; Then, based on the compact-format dynamic linear (CFDL) data model, a discrete sliding mode control algorithm is used to design a data-driven controller structure for crystal diameters, and the stability of the iterative tracking error is verified by the stability analysis; Finally, the proposed MFA-SMILC controller is applied to silicon single crystal diameter control, and compared with the conventional model-free adaptive iterative learning control (MFA-ILC), it is found that MFA-SMILC has faster response speed and convergence speed, which verifies the effectiveness of the proposed control method.