Entropy-Aware Model Predictive Control for Transient Quality Mitigation in Wire-Arc Additive Manufacturing
Md Shafikul Islam, Mahathir Mohammad Bappy, Safiur Rahman Tushar
Pause-induced cooling creates a restart transient in WAAM that electrical signals often fail to reveal. We show that the spectral entropy of the arc sound collapses during this transient and use it as the state of a model predictive controller. In a simulation and surrogate-model study, the entropy-aware controller shortens restart recovery by 47% and reduces the quality-deviation AUC by 35% relative to a thermal-only baseline.
When deposition pauses, the part cools and acts as a heat sink. At restart, current and voltage can remain within their nominal ranges while bead morphology and quality drift. Controllers that rely on electrical or thermal states alone may therefore miss a transient that later appears as a defect.
Treat the arc's acoustic signature as a response-side quality signal. Its spectral entropy H(t) provides a compact, control-ready state. An entropy-aware MPC tracks a reference entropy Href while penalizing abrupt changes in heat input, which yields a short, front-loaded correction.
A microphone records the arc signal x[n] throughout deposition and restart.
A short-time Fourier transform produces the spectrum X(k, t).
H(t) = -Σ pk(t) log pk(t) summarizes spectral spread; it drops sharply after a cold restart.
Learned models (DMDc, SINDy) predict the response; MPC solves q*(t) = argmin [(H - Href)2 + λ(Δq)2].
The optimal input applies a front-loaded heat pulse and relaxes as entropy re-enters the stable band.
All results come from simulation with learned surrogate models, not physical experiments.
Acoustic sensing is inexpensive and non-contact, and it responds to process quality rather than to the commanded electrical inputs. Using it as a control state is a step toward robotic WAAM cells that detect and correct transients during the build instead of relying on post-build inspection.
Md Shafikul Islam, Mahathir Mohammad Bappy, Safiur Rahman Tushar. "Entropy-Aware Model Predictive Control for Transient Quality Mitigation in Wire-Arc Additive Manufacturing." LSU Control Systems Symposium, 2026. doi:10.31390/lsucontrol.26.12.
