Rapid Characterization of Material Heterogeneity through Acoustic Resonance and Physics-informed Artificial Intelligence

TR Number

Date

2026-05-11

Journal Title

Journal ISSN

Volume Title

Publisher

Virginia Tech

Abstract

Metal additive manufacturing is reshaping modern industrial production by enabling the high-throughput fabrication of geometrically complex components with reduced material waste and shortened lead times; as production volumes scale, the ability to rapidly and reliably qualify every as-built component through non-destructive evaluation becomes critically important for ensuring consistent product quality, yet remains an unresolved challenge, particularly in resource-constrained and extreme environments where conventional characterization approaches such as X-ray computed tomography are prohibitively slow and costly for large-volume inspection. To address this gap, this dissertation proposes a novel physics-informed artificial intelligence framework that integrates laser acoustic resonance spectroscopy with physics-based simulation, active learning, Bayesian calibration, and deep learning surrogates to enable rapid, contact-free quality assurance of metal additive manufacturing components. To achieve our research goals, we first establish the theoretical foundation by demonstrating that laser acoustic resonance spectroscopy signals are uniquely sensitive to mesoscale material heterogeneity and by developing an active learning–enhanced framework for bi-directional inference between heterogeneity descriptors and resonance spectra. Subsequently, we extend this framework to industrially relevant additive friction stir deposition conditions through Bayesian calibration of unknown material parameters and a two-stage data augmentation pipeline that generates physically consistent simulation ensembles from a limited experimental dataset. We then characterize a structurally distinct multi-layer specimen cohort deposited on contaminated surfaces, establishing quantitative correlations between mesoscopic flaw features and interfacial mechanical performance that confirm the framework's applicability to realistic manufacturing conditions. Furthermore, we synthesize all preceding components into a fully non-destructive quality assurance pipeline that infers ultimate tensile strength from vibrational measurements alone, achieving cross-process generalization without costly measurement or destructive testing. In conclusion, this dissertation makes a significant contribution to the field of non-destructive evaluation and metal additive manufacturing quality assurance by establishing a new physics-informed paradigm for rapid, low-cost component qualification. Our research findings have direct implications for the industrialization of autonomous manufacturing in aerospace, defense, and space exploration sectors, where reliable and efficient component certification is indispensable for structural integrity and mission success.

Description

Keywords

Metal Additive Manufacturing, Non-Destructive Testing, Quality Control, Bayesian Learning, Deep Learning, Data Augmentation, Finite Element Analysis

Citation