.png){kind=icon}
.png){kind=icon}
Return
Physics-informed online intelligent discharge state identification for high-energy electrical discharge milling: Theory and methods
.png){kind=icon}
PRE
AI
H
K
Z
P
G
P
Z
R
X
纪
DOI：10.1016/j.ymssp.2026.114812.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
High-energy electrical discharge milling (HEDM) offers considerable potential for high-efficiency machining of difficult-to-machine materials in aerospace and other advanced manufacturing sectors. However, compared with conventional electrical discharge machining (EDM), HEDM is characterized by substantially higher single-pulse energy, greater fluctuations in discharge behavior, and lower process stability. To address the limitations of purely data-driven models in terms of physical interpretability and cross-condition generalization, this paper proposes a physics-informed semi-supervised learning framework for online HEDM discharge state identification. First, a time-threshold-based semi-automatic labeling method is developed to generate discharge state labels and identify key pulse-level physical information. Second, an empirical model for estimating the equivalent plasma channel radius is established based on high-energy single-pulse discharge experiments. The estimated radius is then combined with the actual single-pulse input energy and equivalent short-circuit conduction to construct a multi-level physical supervision framework. These three quantities characterize the spatial scale of thermal effects, heat input intensity, and short-circuit-related conduction behavior, respectively. By jointly optimizing the pulse-level classification loss and the window-level physical consistency loss, the proposed framework enables the model to learn discriminative features while constraining its predictions using discharge-related physical quantities. Comparative and ablation experiments demonstrate that the proposed method achieves an identification accuracy of 99.17 ± 0.75% under the new operating conditions and improves both cross-condition generalization and physical consistency. Real-time validation further confirms its capability for online HEDM monitoring, thereby providing a theoretical and technical basis for future intelligent control of HEDM processes.
Keywords:
Physics-informed learning
Discharge state identification
High-energy electrical discharge milling (HEDM)
Semi-supervised learning
Intelligent control
Journal
IF:
8.9
Papers:
1.2W
Citations:
6.6W
Organization
Cited Papers