The growth of digital tools and capabilities within nondestructive testing (NDT) has led to an increased interest in using data-driven methods for analysis of NDT data. These are commonly called artificial intelligence and machine learning (AI/ML). At their core, these methods are based on available data for training and use statistical methods to analyze the data and predict outcomes based on past results. Therefore, a comprehensive analysis of the data quantity and quality is very beneficial to ensure the data being used is sufficient to enable appropriate training of the algorithms. In addition, analysis of the data can provide guidance on the appropriate class of algorithm to use. To illustrate this approach, synthetic data sets of stress intensity factors (SIFs) were evaluated. SIFs model the crack driving force and are a function of geometry, load, crack shape, and crack size. Fourier transform (FT) analysis guided the ML approach by analyzing frequency decay patterns to match data complexity with the most suitable ML model. ML models evaluated include random forest regression (RFR), support vector machines (SVR), neural networks (NN), and deep operator network (DON), The ML models were evaluated as a function of dataset size, resolution, crack scenarios, and hyperparameters on representative SIF values from datasets generated by a high-fidelity finite element model. Quantitative analysis of each ML model determined the mean relative absolute error as compared to ground truth. The results can guide appropriate ML method selection, given data attributes, which include quantity, noise, and complexity, that is relevant for all engineering data. The analysis indicated the need for very large data sets which are extremely challenging to acquire for NDT applications, especially as ML models need to be trained for each individual assessment scenario.

Therefore, enhanced outcomes applied to known NDT data include additional analysis methods, such as heuristics and model-based analysis, to augment the data-driven methods. This approach of combining techniques provided a successful capability to assist NDT inspectors in diagnosing the state of a material based on available NDT data. Representative case studies, including algorithms transitioned into production-based NDT procedures are presented. This will include background on their development and validation process using probability of detection studies. The US Air Force has labeled these approaches as assisted data analysis (ADA). The high degree of variability for typical Air Force applications requires humans to be integrated in the decision-making process for NDT data. Combining the capabilities of all techniques to analyze complex NDT data has been shown to maximize their value in improving the effectiveness, accuracy, and efficiency of NDT assessments that contain a high degree of complexity and variability.

The growing complexity of industrial assets demands nondestructive testing (NDT) methods that provide both reliability and reproducibility. Phased Array Ultrasonic Testing (PAUT) has become indispensable for flaw detection and sizing, yet traditional evaluation remains heavily dependent on inspector expertise, leading to variability in interpretation. With increasing inspection data volumes, the need for automation has never been greater.

This work introduces an automated PAUT evaluation framework that consolidates A-scan, B-scan, C-scan, and S-scan views to deliver comprehensive defect characterization. The methodology employs advanced signal processing with multi-gate analysis calibrated to wall thickness, robust distance-amplitude correction, and hybrid decision logic combining constant false alarm rate detection with cross-correlation. Data fusion across multiple scan modalities strengthens defect evidence, while ridge tracking and scan-shifting enable improved flaw sizing and localization. The inclusion of S-scan enhances angular coverage, ensuring more reliable detection of irregularly oriented defects that may be overlooked in conventional views.

Case studies demonstrate that the proposed workflow reduces operator subjectivity, improves detection reliability, and accelerates interpretation across diverse inspection conditions. The results highlight how automation and multi-view fusion can transform PAUT into a more consistent, data-driven tool that meets the evolving demands of asset integrity programs.

This presentation will benefit inspectors, NDT engineers, and integrity managers by demonstrating practical advancements that align PAUT practice with the future of automated, predictive inspection.

Lithium, a key component in secondary batteries, is increasingly important with the growth of electronic devices and electric vehicles, and is designated as a strategic resource in many countries. Cathode materials such as nickel and cobalt are also rare metals, with extraction processes that cause environmental damage and supply chains vulnerable to external factors. As a result, demand for advanced battery recycling technologies continues to rise. This study examines applying artificial intelligence (AI) to neutron activation analysis data to quantify rare elements in black powder from the recycling process. Monte Carlo simulations and random sampling generated datasets, and a Multi-Layer Perceptron (MLP) model was trained to predict compositional ratios from spectral data. The model achieved low mean squared error values (0.01–0.001), with higher spectral coefficients improving prediction accuracy.
This study was supported by the Commercialization Promotion Agency for R&D Outcomes (COMPA) funded by the Ministry of Science and ICT (MSIT) (No. 2710086067).

The mechanical response of a component is affected by defects arising from the laser powder bed fusion (LPBF) fabrication process, such as porosity. Thus, it is important to develop accurate and efficient inspection methods for identifying porosity. In this work, porosity identified in an X-ray computed tomography (XCT) volume of a Ti-5553 coupon was compared to pores identified in a serial sectioned volume that represented the ground truth. The porosity of the XCT scan was identified using contrast-based, ISO-based, and machine learning (ML) methods for segmentation. Large inherent porosity was easy to identify, but the ISO-thresholding still struggled due to the intensity gradient resulting from both the beam hardening in XCT and the uneven lighting of the serial sectioning panels. The results show that ML based methods were better suited for identifying small pores and reducing the number of false positives. Additionally, high strain-rate impact testing was done on some of the XCT samples as well as post-mortem XCT inspection, and the same suite of segmentation and quantification tools were used to identify the large spallation cavities. The comparison of porosity pre- and post-mortem provides insight on the influence of the LPBF porosity on the formation of spall cavities.

XCT data further enables true quantitative 3D characterization of LPBF components, providing a powerful tool for linking observed defects to mechanical performance. By utilizing XCT-derived volumes, traceable digital twins of components can be generated, capturing the precise geometry and defect distribution of the physical part offering an ideal hybrid environment between measurement and computational framework. These digital twins facilitate predictive modeling of mechanical behavior and defect evolution, that will be necessary for predicting the true direct correlation between experimental observations and computational simulations. This capability underscores the potential of XCT as a critical resource for advancing the application of LPBF processes to critical components, while allowing for an understanding of true defect tolerance in optimizing component design and performance.

ACKNOWLEDGEMENTS:
The work described here was supported by the Additive Coordination Team (ACT) program under the Additive Manufacturing initiative and led by John Carpenter at Los Alamos National Laboratory (LANL) in collaboration with Jonathan Lind at Laurence Livermore (LLNL), Ben Brown at Kansas City National Security Campus, Paul Korinko at Savannah River National Laboratory, and Kevin Shay at Y12. The authors would like to thank Eddie Moffitt and Mathew Dennis for characterizing with computed tomography and Caleb Schauble for ultrasonic imaging. We would also like to thank the team at the Dynamic Integrated Compression Experimental (DICE) facility at Sandia National Laboratories (SNL) for preparing targets and executing the dynamic and Thor experiments. Finally, the staff at HP-CAT for their support in diamond anvil cell experiments at Argonne National Laboratory.