The paper highlights a new paradigm using simulation-based analysis that employs physics-based models that use parallel processing using GPU for rapid generation of synthetic data sets for trainings AI engines for the purpose of ADR. This simulation tool coupled with a mechanical motion based simulation toolkit allows for the creation of digital twins for NDE systems.
For the development of a simulation assisted ADR (Automatic Defect Recognition) algorithm (SimADR) that is based on Deep Learning (DL) and/or Machine Learning (ML) mode, our approach uses limited experimental/field NDE/NDT data sets that are available and by deriving critical statistical distribution parameters from this data set, the stochastics of the simulation models are determined. Thereby, the simulated data sets are generated using the numerical simulations along with the variations in the different parameters during experimental/field data acquisition. This process allows the generation of simulated data sets in large quantity that augments the smaller data sets obtained experimentally. This rich data set is subsequently utilized to train the DL models and provide reliable ADR algorithms. Weld data sets from Digital X-ray Images and PAUT (with FMC/TFM) are both used to demonstrate the SimADR approach.
Digital twins are an essential part of Industry 4.0 where digital models are used to mimic the physical systems and to enable seamless interactions between the digital and physical system in an interactive manner. The use of Digital Twins for NDE 4.0 would involve the reproduction of the NDT automated systems in the digital world that would incorporate the mechanical robotics along with the physics of the NDT technique to provide a simulation platform. This Digital Twin will enable the virtual inspection protocols to be evaluated, without the need for the access to the physical system and optimize the inspection prior to porting the optimized parameters to the physical system for inspection. Alternatively, the data from the physical system can be used by the digital twin to emulate the inspection and provide real-time guidance to the inspector. In this talk, the generation of digital twins for ultrasonic gantry-based inspection system for large composite structure with complex shape will be demonstrated. Additionally, a digital twin for an X-ray digital flat panel-based system for inspection of Aluminum cast components will also be discussed. The digital twins use the physics of the ultrasonic wave and X-ray interaction with the component and the simulation of the actual datasets obtained from the simulations along with the manipulation of the components.

Industrial image data for inspection collected by nondestructive testing (NDT) in production environments is inherently diverse. Different products, exposure settings, imaging systems, signal-to-noise ratios (SNR), material thicknesses, and weld geometries can vary significantly across different sites. Despite these challenges, most existing defect detection models using artificial intelligence assume a uniform data distribution, leading to decreased performance when applied to heterogeneous data. To address this issue, we present a defect-detection framework based on Normalizing Flows (NF) specifically designed for heterogeneous image data. Instead of using the standard single-Gaussian prior, we employ a Mixture of Gaussians (MoG) in the latent space in which each component of this mixture captures a unique visual mode. We introduce two auxiliary loss terms: a cluster separation loss and a consistency loss. The cluster separation loss promotes the disentanglement of latent features from different sources and modes, thereby aiding the isolation of defect-related regions. The consistency loss stabilizes the model against minor perturbations in noise, brightness, or texture. Experimental results on real-world datasets demonstrate the effectiveness of our approach and its potential for practical deployment in NDT workflows.

This research introduces a handheld spherical non-destructive testing (NDT) probe that integrates a PVDF/Graphene (PVDF/Gr) piezoelectric membrane into a spherical probe. The instrument reconstructs 3D surface geometry using Simultaneous Localization and Mapping (SLAM)-referenced six-degree-of-freedom (6-DoF) localization. Unlike conventional instruments such as rollers or flat probes, this work proposes a spherical-shaped probe with piezoelectric membranes distributed over its entire surface. This design enables surface geometry mapping while eliminating the orientation constraints that typically occur with roller-type probes during scanning. Each piezoelectric membrane is arranged as several tiles and labeled according to its geodesic position on the sphere, so the location of the piezoelectric signal generated when contact occurs between the tiles and the inspected surface can be identified with respect to the spherical shell. A non-contact inductive coupling mechanism embedded in the rotating spherical assembly enables the piezoelectric signals to be transferred wirelessly, allowing rotation without slip or cable-induced constraints. The sensing instrument integrates with IMU and gyroscope to continuously estimate the global 6-DoF coordinate of the probe. The system computes the location of each sensor tile in global coordinates and infers the surface contact point along the outward normal vector of that tile. This produces a real-time mapping pipeline in which the inspected field is represented as a 3D distribution of positions derived from the piezoelectric responses. The spherical-shaped probe facilitates inspection of curved, irregular, or partially occluded surfaces. In addition, the distribution of piezoelectric tiles over the entire probe surface enables mapping of surface texture, irregularities, and material compliance in the inspected region. Spherical shape provides a consistent contact behavior in any moving direction. It is allowing the probe to roll continuously over a complex curvature surface. The experimental results will be presented to show the different signals generated by the spherical PVDF/Gr probe scanning through surfaces with various surface characteristics: smooth, rough, weld geometries, corroded, uneven, etc. This research highlights an innovative approach to portable NDT instruments by integrating piezoelectric sensing with spatial mapping. Such an innovation is well suited for digital-twin updates, surface degradation monitoring, and automated defect interpretation. The spherical PVDF/Gr probe represents a significant advance toward practical handheld systems that integrate piezoelectric intelligence with 3D mapping for next-generation inspection workflows.

Abstract

Industry 4.0 and NDE 4.0 had the promising premise for a place where all data would be speaking the same language. To make this promise a reality, an open file format for Non-Destructive Evaluation (NDE) acquisitions is crucial as it offers data sets that are readable and interpretable without proprietary software or software development kits, software interoperability eases exchanges, and readability is persistent.

An open file format is proposed as an alternative to files that require specific proprietary applications to load, read and save. The organization of both acquisition data and configuration information is supported by a Hierarchical Data Format (HDF5) which further provides extensive data storage capacity and readability by various software and programming languages. An accompanying documentation defines the various fields of the file. Some examples of currently available use cases are presented, including common phased array angle beam weld inspection and corrosion screening of plates and tubes. The new file format being highly adaptable, the format will support more demanding acoustic inspection use cases in the future without producing breaking changes in the generic field definition. The file format is also a multi-technology format, and support for eddy current technology as well as other NDE technologies is planned in coming releases. Overall, the open file format provides an all-in-one versatile and highly sharable alternative to proprietary files that promotes both a clear and robust description of the data.

The proposed open file approach contributes to NDE 4.0 by reducing barriers with an open interface making NDE inspection data more available for customized analysis applications as well as emerging themes such digital twins, statistical analysis, and artificial intelligence.

Keywords: NDE 4.0, open file format, interoperability, UNIS, phased array ultrasonic testing (PAUT)