Chien-Ping Chiou

Modeling of Ultrasonic Signals from Weak Inclusions

Recent research efforts aimed at improving the detection of hard-alpha inclusions have emphasized the need for accurately modeling the responses from such weakly-reflecting inclusions. The need arises because of the rare natural occurrence of hard-alpha inclusions, and consequently, the lacks of suitable experimental samples. These difficulties lend impetus to the application of signal modeling to augment and extend the experimental data in assessing detectability. Currently, a new approach is being developed for the purpose of predicting time-domain echoes from inclusions of specified morphology. This work is the continuation of our previous study of flat-bottomed holes [1–2] in constructing a methodology for estimating the probability of detection of flaws in titanium alloys based on a combination of physical and statistical models.

New approaches to model‐based ultrasonic flaw sizing

Ultrasonic equivalent flaw sizing methods have been recently used to size a flaw in a material by obtaining a best‐fit simple equivalent shape that matches the ultrasonic scattering data. However, current ultrasonic equivalent flaw sizing methods have a number of important limitations: (1) they are both iterative and highly nonlinear in nature, and (2) they require the availability of flaw classification information. Here, a series of approaches are outlined that address both of the above problems. Both numerical and experimental results that validate these new approaches are also given. First, when the flaw shape is determined in terms of a best‐fit equivalent ellipsoid, a new linear least‐squares/eigenvalue method is described that can replace existing nonliner routines and provide a computationally fast and robust sizing procedure. Second, it is shown that if the shape is determined instead in terms of an expansion in spherical harmonics, the sizing problem again reduces to a simpler linear least‐squar...

Ultrasonic Signal Characterizations of Flat-Bottom Holes in Titanium Alloys: Experiment and Theory

The POD Working Group under the Engine Titanium Consortium [1] is currently conducting a series of ultrasonic experimental and theoretical investigations using aircraft engine titanium alloy specimens. These efforts, in conjunction with corresponding statistical model developments, represent a new attempt to construct an integrated physical/statistical methodology for estimating the probability of detection (POD) of flaws in (titanium) materials. An important aspect of this new methodology is the use of physical models, to the extent possible, to predict the flaw and noise signals under the influence of various inspection parameters, thereby reducing the experimental effort and providing a basis for extrapolating to cases not covered by experiment. The ultrasonic investigations described in this paper are required to validate the accuracy and range of applicability of the models before they can be utilized with confidence.

Development of Ultrasonic Models for Hard-Alpha Inclusions in Titanium Alloys

This paper describes research directed towards modeling ultrasonic signals from hard-alpha inclusions in titanium alloys. The modeling effort has been made difficult by the complicated morphology of such inclusions which can include voids, cracks, core and diffusion zones. Fortunately, a large portion of hard-alpha inclusions are acoustically weak scatterers in nature, and advantage can be taken of simplifications as afforded by Born approximation and some ad-hoc interface conditions. Models along these lines have been previously developed and their validations on synthetic hard-alpha inclusions of cylindrical shape at normal incidence have been reported [1]. Extensive use of these ultrasonic models were also presented in the development of a statistical methodology for estimating the probability of detection [2,3]. In current work, we extend the model capability to include arbitrary flaw orientation and oblique incidence. Model predictions are compared with experimental data collected from titanium specimens for different beam angles, focal depths, inclusion sizes and orientations. The range of the model applicabilities and their possible extensions will be presented. Morphological modeling of the three-dimensional, naturally-occurring inclusions based on stacks of two-dimensional metallographic measurements are also described.

Ultrasonic Flaw Detection Using Neural Network Models and Statistical Analysis: Simulation Studies

Flaw detection problems in ultrasonic NDE can be considered as two-class classification problems, i.e., determining whether a flaw is present or not present. To be practical, a flaw classification method must be able to handle the uncertainties associated with interference from grain noise which leads to poor signal-to-noise ratios (SNR). In this work, the use of neural network models and statistical correlation is demonstrated for one such detection/classification problem. In particular, based on simulation studies, we wish to establish practical strategies in detecting weak volumetric flaw signals corrupted by high grain noise. An example of this type that is of recent interest is the detection of “hard-alpha” inclusions in aircraft titanium components [1]. Both the feasibility and reliability of using these classifiers are assessed. This effort was carried out in parallel with another study [2] where more traditional signal processing approaches were taken.

Ultrasonic Flaw Detection Using Signal Matching Techniques

Detection of hard-alpha inclusions in titanium has been a challenging problem for over two decades. Hard-alpha inclusions are brittle regions of microstructure usually resulting from oxygen or nitrogen contamination. During the high-stressed manufacturing process, these regions initiate cracks which are likely to grow during the service of the component, possibly leading to its failure. It becomes imperative, therefore, to detect these regions early in the manufacturing process. The detection, however, is compounded by the small contrast (consequently weak ultrasonic signal strength) of these inclusions, and the presence of high-level, correlated grain noise with spectral characteristics similar to hard-alpha inclusions. Earlier studies [1] based on model-generated simulation data have suggested that signal matching techniques are promising candidates for the detection of hard-alpha inclusions. One of the primary advantages in the use of these techniques lies in their ability to use flaw signals obtained by ultrasonic modeling as promising filter kernels.

Model-Based Signal Processing Techniques for Ultrasonic Flaw Detection: Simulation Studies

The ultrasonic signals observed in inspection processes can often be accurately predicted by suitable measurement models. These model predictions can be used to provide important information to guide the development of subsequent signal processing algorithms. Here such a hybrid use of ultrasonic modeling and signal processing is demonstrated in the context of the problem of detecting ultrasonic flaw signals in noise. In particular, we wish to apply this hybrid methodology as an initial approach to solving the problem of detecting hard-alpha inclusions in titanium alloys.

Development of Geometrical Models of Hard-Alpha Inclusions for Ultrasonic Analysis in Titanium Alloys

The engine titanium consortium is currently conducting an extensive study of the ultrasonic response and detectability of a number of naturally occurring hard-alpha defects found in titanium billets. These naturally occurring defects appear to be highly irregular in shape and inhomogeneous in composition, and we would like to access the extent to which their responses can be predicted by available ultrasonic scattering models [1]. Three dimensional geometrical (surface and solid) models of the hard-alpha defects are needed in order to obtain the geometrical and material properties to drive the ultrasonic model calculations and the subsequent probability-of-detection evaluation [2].

A NEURAL NETWORK MODEL FOR ULTRASONIC FLAW SIZING

Abstract A multilayered neural network model is used here for solving an inverse sizing problem in the field of ultrasonic nondestructive evaluation. In particular, a feed-forward network trained via the error backpropagation algorithm is shown to be able to invert size and orientation information for circular cracks from time domain ultrasonic data. Test results of the networks performance on both theoretical and experimental data are presented. A new adaptive learning scheme for improving the training speed of such methods is also presented.

Ultrasonic Flaw Classification — An Expert System Approach

An expert system, FLEX, for classifying isolated flaws as either crack-like or volumetric has been under development at the Center for NDE, Iowa State University. Previously, we have described the overall design of the system [1], which is composed of two cooperating systems FEAP and FLAP. The feature processing (FEAP) system is designed to extract fundamental features in the ultrasonic signals that are indicative of cracks or volumetric flaws. The flaw processing (FLAP) system then uses the existence (or non-existence) of these features to classify the flaw. FLAP is structured as a classical rule-based expert system and has also been described previously [2]. Here, we will present the major elements of FEAP and the design philosophy that has gone into its construction. A more detailed account of FEAP is given in the thesis of Christensen [3].