Course Agenda

Overview of USAF Perspectives on NDE Imaging, POD, and Characterization

Eric Lindgren

Air Force Research Laboratory, Wright-Patterson Air Force Base, Dayton, OH, USA

Outline of talk:

- Introduction to NDE and how it is used within the USAF for sustainment and new materials quality control / certification

- Rapid review of USAF POD guidance (MIL HDBK1823)

- Need for standardized capabilities in the USAF

- How we standardize capability

- Image-based methods used within the USAF

- Current methods to validate (e.g. POD) image-based methods

- Overview of current characterization R&D supported by AFRL

- How we intend to validate characterization methods (i.e. POD equivalent for characterization)

- Short review of MAPOD to set stage for extended talk at conference

- Summary

Imaging, Sizing and POD with ultrasound – challenges which have to be overcome

Thomas Heckel, Marija Bertovic, Mate Gaal, Rainer Krankenhagen; BAM, Berlin, Germany

Ralf Holstein; DGZfP, Berlin, Germany

Abstract: Each inspection task poses its own specific challenges to imaging, sizing and classification of indications and well as probability of detection (POD) estimation. This statement is valid in particular for the ultrasonic inspection of components with complex geometries or components made of complex materials or even both. Fiber reinforced plastic (FRP) is one of the most common examples of complex materials inspected by non-destructive testing methods. Applications of FRP composites in modern industries are increasing due to their considerable advantages such as light weight and excellent mechanical properties. Defects in FRP structures may be introduced during the processing and fabrication of composite components and can initiate or grow in-service. In order for a particular NDE technique to achieve broad acceptance by industry, it is desirable for the technique to be able to detect a range of defect types with a high level of confidence. For ultrasonic inspections of these parts standard techniques, phased array techniques and air coupled techniques are available and solutions will be presented and compared against each other. Methods for data analysis will be discussed. A good example of an inspection of parts with complex geometry is the inspection of railway axles. These axles are safety relevant parts and therefore have to be inspected regularly. For in-service inspections on hollow wheelset axles, manual or mechanized ultrasonic inspection systems with a rotating ultrasonic probe are used to inspect the outer surface for flaws. The surfaces of these flaws are typically positioned in the radial-radial plane perpendicular to the axis of the axle. The inspection is performed with a helix like movement of the beam along the testing zones. At DGZfP training centre Wittenberge/Germany round robin tests with the trainees have been carried out on three hollow axels featuring three flaws with different depth and positions each to statistically evaluate the overall performance of the inspection. The evaluation was focused on the determination of position and size/amplitude of the flaws at different locations of the complex geometry of the axle to obtain information about the probability of detection (POD). Results of this study will be presented. A reliability analysis, or any analysis of the capability of NDT to achieve its purpose—whether it’s to detect, characterize or classify flaws—is incomplete without the consideration of human factors. Whereas POD provides information about the system’s capability under controlled laboratory conditions, studying human factors provides an insight into the reliability and likelihood of failure in the field. How the knowledge of human factors, i.e. various factors affecting NDT inspectors in the field, can be used to optimize NDT reliability, productivity, inspection process and personnel training will be demonstrated.

Simulation Assisted Automation in Defect Recognition Intelligence for Radiography Data Sets

Krishnan Balasubramaniam

Indian Institute of Technology Madras, Chennai, India

Abstract: In this presentation, the model based approach using numerical models of X-ray imaging forms the basis for the development of a simulation assisted ADR (Automatic Defect Recognition) algorithm that is based on Deep Learning (DL) mode. The approach addresses the classic issues during the implementation of DL approach to industrial applications, such as X-ray data interpretation, that includes lack of sufficient data as well as biases in the data sets, among others. Here, using the limited experimental/field 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. Industrial applications will be used to illustrate the feasibility of this technique.

The Next Generation of Ultrasonic Flaw Sizing Methods – One Perspective

Lester W. Schmerr Jr.

Department of Aerospace Engineering, Iowa State University, Ames, IA, USA

Abstract: Flaw sizing with ultrasonic waves is currently done primarily with either the dB drop method or the time of flight diffraction (TOFD) method. Both of these legacy sizing methods have known limitations. With the growing use of phased arrays, phased array imaging has also been used for sizing purposes. Here we describe a possible next generation of flaw sizing methods based on the concept of equivalent flaw sizing and the use of phased arrays. Equivalent flaw sizing uses the same type of input data as TOFD to size isolated cracks but it requires more of that data to enable the extraction of size parameters in terms of a best fit elliptical-shaped crack. It is demonstrated that the scanning capability of phased arrays can be used to help collect the needed data - data which can then be used in a simple two-step sizing method to extract size information in a form that is usable in fracture mechanics studies. It is shown that equivalent flaw sizing methods in various different forms also can be used in principle to size surface-breaking cracks and volumetric flaws. Equivalent flaw sizing with phased arrays does not use the imaging capabilities of those arrays. When phased array imaging is used for sizing purpose it is important to understand the image formation process when using current popular imaging methods such as SAFT and TFM. Here, we also discuss the physics behind those imaging methods that can help to prevent their misuse for sizing purposes.

Taking nonlinear ultrasound from the lab to the field – the importance of developing reliable standards and benchmark problems

Laurence Jacobs

College of Engineering, Georgia Institute of Technology, Atlanta, GA USA 30332-0360

Abstract: Predictive health monitoring of structural components requires the development of advanced sensing techniques capable of providing quantitative information on the damage state of macroscale structural systems. Nonlinear ultrasound (NLU) can measure absolute, strength based material parameters which can be coupled with uncertainty models to enable accurate and quantitative life prediction. These second harmonic techniques are acoustic wave based, so component interrogation can be performed in situ using either bulk, surface and guided waves. Since the underlying material physics is independent of the choice of ultrasonic wave, it is possible to interrogate a variety of different structural systems for microscale defects. Recent studies have demonstrated that NLU is sensitive to nano- and microscale changes in microstructure due to a variety of damage mechanisms such as thermal, sensitization and fatigue damage. Although linear ultrasonic methods are effective in detecting macroscopic flaws such as cracks, they are often not capable of assessing damage in its early states. In contrast, NLU is sensitive to damage mechanisms associated with nanoscale changes in microstructure, such as dislocations and precipitates. This talk will examine the steps needed to take NLU from being a valuable laboratory characterization technique to a reliable, field qualified tool

Extracting information for NDE via tomography with x-rays and ultrasound

1. Huthwaite, Imperial College London, UK
2. A. Jones, Swansea University, UK

Abstract: Non-Destructive Evaluation utilises measurements to extract information about the underlying state of a component. In imaging and tomography approaches, multiple measurements are combined to extract information which would not otherwise be available, opening up a range of new possibilities. This talk will explore this concept of information extraction within both x-ray tomography and guided wave tomography. In the case of x-ray tomography, the focus will be on generating images from reduced datasets for speed and cost benefits, and in guided wave tomography the goal will be to utilise improved understanding of the physical guided wave behaviour. In both cases, there is an emphasis on maximising the information that can be derived from the measured data.

Model-based computational tools for imaging, diagnostics and POD estimation

Pierre Calmon and coll.

CEA LIST, France

Abstract: The role of modelling through simulation package for the design and the performance demonstration of inspections is well established. Over the years, efforts are put on forwards models in order to accompany the evolution of techniques and practices. Moreover, this evolution create new needs for a variety of model-based tools and motivate new modelling strategies and new paradigms. On one hand, inspection techniques give access to tremendously increased amount of data. It motivates the development of model-based imaging and automated diagnostics (identification and sizing of defects) tools with a real-time objective. The recent progresses in the field of machine-learning and artificial intelligence offer new ways to reach these goals. On the other hand, the demand for a comprehensive handling of uncertainties in the process of performance demonstration (reliability/accuracy) implies intensive computations coupled to statistical studies. In this talk we will give an overview of recent works achieved at CEA LIST in the aim of proposing computational tools fulfilling these needs. The presentation will principally focus on ultrasonic and guided waves NDE and SHM techniques. In a first stage, we will propose a glimpse on the recent capabilities of forwards numerical models and will discuss modelling strategies. We will introduce the complementarity of data-driven and physic-based models and emphasize on metamodeling strategy which appears to be quite powerful for the different objectives listed above. Then, we will illustrate the application of these tools in the context of NDE and SHM reliability assessment. Finally, we will discuss recent works aiming at proposing enhanced diagnostics capabilities. Beyond the development of (possibly adaptive) model-based imaging techniques we will show how inversion and machine learning can be used for classification and characterization of defects.

META-MAPOD: Open-Source Framework for Metamodel-Assisted Probability of Detection

Leifur Leifsson

Department of Aerospace Engineering, Iowa State University, Ames, IA, USA

Abstract: The practice of determining the probability of defect detection for a NDE inspection through use of computational measurement simulation, so-called model-assisted probability of detection (MAPOD), is recognized for its potential cost savings and improved accuracy. Measurement simulation covering the full range of possible input parameters most often represents a prohibitively large computational task, particularly when using rigorous high-accuracy numerical simulation of the measurement physics. Metamodels (also called surrogate models) can be used in lieu of the rigorous computational physics models to improve the efficiency of the MAPOD analysis. The metamodels are derived from results obtained by running the rigorous simulations for a substantially smaller, judiciously chosen parameter matrix. The accurate yet computationally efficient approximation of simulation results enables the full exploration of the parameter space in evaluating the POD. In this talk, an open-source framework for efficient metamodel-based MAPOD analysis will be described. Specifically, the theory, structure and capabilities of the framework will be presented, and results of applications to several analytical and MAPOD examples will be given.