Multimodal Digital Twin with Uncertainty-Aware Temporal Fusion for Adaptive Electrical Discharge Machining Control

Electrical discharge machining (EDM) is governed by transient, stochastic discharge phenomena that are difficult to characterize with a single sensing channel or a purely data-driven model. This paper proposes a conceptual physics-informed multimodal Digital Twin framework for uncertainty-aware monitoring and adaptive control of EDM. The framework couples modality-specific encoders for electrical, acoustic, vibration, and thermal signals with a temporal attention-based fusion mechanism to build a latent process-state representation constrained by established discharge-energy and thermal relationships. This latent state continuously updates a Digital Twin, which propagates the estimate forward through a multi-step predictive model to anticipate future material removal rate, surface quality, and failure risk, while explicitly quantifying predictive uncertainty rather than issuing point estimates alone. A risk-aware model predictive control layer evaluates candidate parameter adjustments against the Digital Twin’s uncertainty-weighted predictions before they are applied to the physical process, so that the controller can act conservatively when confidence is low. The contribution of this work is the architecture and mathematical formulation of this integration rather than an experimentally validated implementation; no datasets, experiments, or numerical results are reported. The framework is intended to guide future data collection, model development, and closed-loop validation studies on physical EDM platforms operating under variable machining conditions