Selected work

Representative Papers

Bond strength uncertainty quantification via confidence intervals for nondestructive evaluation of bonded composites

Dec 15, 2025

In ultrasonic non-destructive evaluation of aerospace composite bond strength, large uncertainties and unreliable confidence intervals hinder safety-critical assessments. To address this under limited-sample conditions, we propose an optimized confidence interval construction method. Our approach integrates a nonlinear forward model with unknown-variance estimation to establish a novel stiffness-to-strength uncertainty propagation paradigm; incorporates coverage calibration to guarantee nominal coverage probability; and combines swept-frequency ultrasonic phase analysis with statistical inverse modeling for robust interval regression. In multi-noise simulation experiments, our method significantly improves confidence interval coverage—particularly under high-noise and parameter-boundary conditions—while simultaneously reducing interval width. The resulting intervals provide interpretable, verifiable, and quantitatively rigorous evidence for bond strength safety validation.

0 citationsRead paper

Generative Modeling of Microweather Wind Velocities for Urban Air Mobility

Mar 04, 2025

Urban Air Mobility (UAM) faces significant safety risks due to the high spatial heterogeneity and stochasticity of local microscale wind, which are difficult to monitor in real time. To address this, we propose a lightweight, probabilistic generative model that efficiently maps coarse-grained meteorological forecasts to high-resolution, spatiotemporally resolved microscale wind fields. Methodologically, we introduce the first integration of Denoising Diffusion Probabilistic Models (DDPM), Flow Matching, and Gaussian Mixture Models (GMM), conditioned on SoDAR-measured microscale wind data and co-located macro-scale weather forecasts—requiring only temporary on-site sensing, with no permanent infrastructure. Experiments demonstrate substantial improvements over conventional deterministic approaches, particularly in turbulence representation, short-term wind variability capture, and physical plausibility. The model enables deployable, real-time wind risk assessment for UAM operations.

0 citationsRead paper

Recent publications

Latest Papers

Bond strength uncertainty quantification via confidence intervals for nondestructive evaluation of bonded composites

Dec 15, 2025

In ultrasonic non-destructive evaluation of aerospace composite bond strength, large uncertainties and unreliable confidence intervals hinder safety-critical assessments. To address this under limited-sample conditions, we propose an optimized confidence interval construction method. Our approach integrates a nonlinear forward model with unknown-variance estimation to establish a novel stiffness-to-strength uncertainty propagation paradigm; incorporates coverage calibration to guarantee nominal coverage probability; and combines swept-frequency ultrasonic phase analysis with statistical inverse modeling for robust interval regression. In multi-noise simulation experiments, our method significantly improves confidence interval coverage—particularly under high-noise and parameter-boundary conditions—while simultaneously reducing interval width. The resulting intervals provide interpretable, verifiable, and quantitatively rigorous evidence for bond strength safety validation.

0 citationsRead paper

Generative Modeling of Microweather Wind Velocities for Urban Air Mobility

Mar 04, 2025

Urban Air Mobility (UAM) faces significant safety risks due to the high spatial heterogeneity and stochasticity of local microscale wind, which are difficult to monitor in real time. To address this, we propose a lightweight, probabilistic generative model that efficiently maps coarse-grained meteorological forecasts to high-resolution, spatiotemporally resolved microscale wind fields. Methodologically, we introduce the first integration of Denoising Diffusion Probabilistic Models (DDPM), Flow Matching, and Gaussian Mixture Models (GMM), conditioned on SoDAR-measured microscale wind data and co-located macro-scale weather forecasts—requiring only temporary on-site sensing, with no permanent infrastructure. Experiments demonstrate substantial improvements over conventional deterministic approaches, particularly in turbulence representation, short-term wind variability capture, and physical plausibility. The model enables deployable, real-time wind risk assessment for UAM operations.

0 citationsRead paper