Ultrasonic power adaptive adjustment for ultrasonic assisted die-sinking electrical discharge machining based on deep deterministic policy gradient

Abstract

Ultrasonic assisted electrical discharge machining (UAEDM) has gained traction, with numerous investigations demonstrating its potential to enhance machining efficiency and surface quality under specific conditions. Ultrasonic vibration is widely recognized for increasing the inter-electrode flow rate, which improves the removal efficiency of discharge debris. However, the inherent complexity of the discharge process makes it difficult to establish a precise mathematical relationship between discharge states and ultrasonic vibration. Consequently, research on the adaptive adjustment of ultrasonic power (amplitude) based on real-time measurement and discharge state statistics throughout the entire machining process remains limited. This paper presents a novel ultrasonic power adaptive adjustment system for die-sinking EDM, leveraging online measurement and deep deterministic policy gradient (DDPG), a state-of-the-art deep reinforcement learning (DRL) approach. By harnessing the powerful nonlinear mapping capabilities of deep learning, the proposed system uses two-dimensional discharge signals measured online as the model input to dynamically modulate ultrasonic power. Experimental results demonstrate that the DDPG-based adaptive system outperforms the fixed ultrasonic power method, achieving improved discharge continuity. Moreover, the servo system significantly reduces retraction servo commands, minimizing reverse motion of the servo system and thereby enhancing overall machining efficiency.