Chatter suppression in nonlinear milling of a flexible plate-workpiece with attached piezoelectric actuators: Comparison of soft-actor-critic-based controller vs optimized type-2 fuzzy controller

• Milling process including cutting force nonlinearities, tool wear & process damping. • Chatter suppression using piezoelectric patch actuators during the milling of a nonlinear flexible plate-workpiece. • Designing a type-2 fuzzy controller for chatter suppression & optimizing its parameters with Genetic algorithm . • Designing a controller based on deep reinforcement learning soft-actor-critic (SAC) algorithm for chatter suppression. • Comparison of SAC-based & optimized type-2 fuzzy controllers via stability lobes, vibration responses & power spectrums . Milling of flexible and thin workpieces is widely used in the industry, but traditional tool control is ineffective. The innovation of this paper is in realistic theoretical methods and explores chatter suppression in the milling of a large, low-frequency workpiece through the application of piezoelectric patches and a reinforcement learning-based controller, which has not been employed before. Due to uncertainties and nonlinearities in the milling process, two model-independent controllers are presented: one based on deep reinforcement learning (DRL), employing the soft-actor-critic (SAC) algorithm, which is compared with an optimized type-2 fuzzy controller to analyze the effectiveness of the new SAC method. Initially, a developed nonlinear model based on experimental data for milling cutting forces is presented to serve as an accurate model for agent training. Nonlinear equations for a cantilever plate with attached piezoelectric patches and accelerometers are derived using von Karman strains, Hamilton’s principle, and the constitutive equations of piezoelectric patches. The coupled nonlinear delayed differential equations (NDDEs) are obtained through the Galerkin and mode summation method. Furthermore, tool wear and process damping (PD) effects are included in these equations. A type-2 fuzzy controller is designed, and its parameters are optimized with a genetic algorithm. Subsequently, an agent is trained using the SAC algorithm of the deep reinforcement learning method for chatter suppression. Eventually, the performance of these two methods is compared in several factors, including the stability lobes, vibration and power spectrum responses, control voltages, and efficiency after the occurrence of chatter, considering the voltage saturation and sampling time. The SAC-based controller outperforms the optimized type-2 fuzzy controller, demonstrating superior material removal rate (MRR), robustness, and optimal responses under varying conditions, proving its effectiveness in machining vibration control.