As advanced materials are introduced into aircraft engines, operating under higher temperatures at greater stresses for longer lifetimes, the need to detect small, internal defects becomes increasingly important. Ultrasound is often the preferred inspection tool because of its ability to penetrate to the interior of a component. However, sound energy reflected from microstructural features in the component produces a background inspection "noise" which is seen even when no defects are present. This noise can inhibit the detection of sound energy reflected from critical internal defects such as cracks, pores, or inclusions.

The primary objectives of this work are to develop a quantitative understanding of the factors which influence the detectability of small defects in advanced engine alloys, and to lay the foundation for an engineering methodology to predict detection capabilities. To focus the work we concentrate on the specific problem of ultrasonically detecting "hard-alpha" inclusions in titanium aircraft engine alloys. These inclusions result from excess local concentrations of oxygen or nitrogen which occasionally occur during processing. Such impurities tend to occupy interstitial sites and cause excess brittleness. An engineering understanding of their detectability requires three elements: (1) knowledge of the strength of the backscattered noise signals associated with normal microstructural inhomogeneities such as grain and phase boundaries; (2) knowledge of the strength of the competing ultrasonic signal reflected by the inclusion; and (3) the use of this information to predict quantities which bear on the probability of detection. Major progress was made in each of these three areas.

In the course of the project, three models were developed for the prediction of absolute noise levels in normal-incidence, pulse/echo, ultrasonic immersion inspections. These are identified as the Independent Scattering Model for Tone Burst pulse inspections (ISMTB), the Independent Scattering Model for Broad Band pulse inspections (ISMBB), and the Monte-Carlo noise Model (MCM). Each model assumes that the backscattered noise is primarily due to single scattering by the individual grains in the metal specimen, and the models consequently apply to low-noise materials. The observed absolute noise level in a given setting will depend upon the microstructure of the specimen, and upon the details of the measurement system (e.g., the transducer and pulsing unit used, the inspection waterpath, the amplifier gain settings, etc.) Both types of dependencies are incorporated into our models. In the ISMTB and ISMBB, the dependence on microstructure enters primarily through a Figure-of-Merit (FOM) for inherent noise severity, which is a property of the specimen alone and is determined by the density of grains and the average scattering capability of a single grain. These two models relate the FOM and measurement system parameters to average noise characteristics, such as the position-averaged root-mean-square (rms) noise level. Either the ISMTB or ISMBB can be employed in two distinct ways: to deduce the FOM of a specimen from measured noise signals; or to predict average absolute noise levels for various inspection scenarios when the FOM is known. For the MCM, the microstructural inputs are more detailed, but predicted noise properties are more detailed as well: e.g., both peak and average noise levels can be estimated.

In our report we document the underlying assumptions and mathematical development for each noise model, and we report on extensive experimental studies carried out to validate the models. For single-phased, equiaxed, randomly-oriented metals it is possible to estimate the FOM from photographs of the microstructure and knowledge of the elastic constants. The FOM value so obtained can be directly compared to that deduced from our model-dependent analysis of backscattered noise. The two values are generally found to agree to within a factor of 2. This level of agreement is considered to be quite good since the noise model contains no adjustable parameters and the predicted average noise level is typically 50-60 dB below a measured front-surface "reference" signal. For two-phase commercial titanium alloys, it is not yet feasible to determine the FOM from microphotographs and related information. However, it is straightforward to determine the FOM by analyzing backscattered noise data. In such cases the deduced FOM is found to be approximately independent of the measurement system parameters, as expected. However, the FOM has been found to vary significantly from specimen to specimen in a suite of commercial alloys, and with direction within a given specimen. These variations are believed to originate from "macrostructural" details related to the processing history of the specimen. We describe how particular etching and photographic methods can be used to reveal this macrostructure, and how supporting data can be obtained from x-ray diffraction studies.

To estimate the strength of ultrasonic signals reflected from hard-alpha inclusions, one requires a knowledge of how the elastic moduli and density of the inclusions differ from those of the host alloy in which they reside. Based on reviews of the literature plus additional experiments conducted as needed, the influence of interstitial oxygen and nitrogen on those properties has been determined. It is clear that, at solute concentrations of a few percent, there are sufficient changes in the moduli to produce significant ultrasonic signals from hard-alpha inclusions in single-phase microstructures. In two-phase titanium alloys, the hardening of the alpha phase can be accompanied by a conversion of the beta phase to alpha. When it occurs, this conversion may diminish the impedance difference between the inclusion and host metal, and consequently make ultrasonic detection more difficult. For specific, reasonable choices of hard-alpha properties, we have used previously developed models to predict absolute defect signal amplitudes for a range of inclusion diameters. These have been combined with noise model predictions to obtain estimates of signal-to-noise (SIN) ratios for hard-alpha inclusions in representative titanium alloys. Such calculations were performed for hypothetical inspections using both focussed and planar transducers. The calculations indicate that the SIN ratio is approximately inversely proportional to the width of the incident sound beam in the vicinity of the defect. Thus, defect detection can be substantially improved by properly focussing and scanning the beam in the interior of the component being inspected. In addition to performing illustrative SIN calculations, we have developed approximate formulas which allow rapid estimation of relative and absolute SIN ratios. These can be used to estimate the optimal choices of transducer diameter, focal length, and waterpath for inspecting a given region of the component's interior.

Much of the formalism developed in this work is fairly general in scope, and is consequently applicable to a wide range of defect-detection problems. Our noise models and associated formulas for SIN ratios can be readily extended to the case of normal-incidence inspection through a curved water/metal interface, and such extensions are currently in progress under different sponsorship. These developments will allow one to estimate SIN ratios for simulated inspections of cylindrical metal billets, and hence to determine the optimum inspection parameters. Straightforward extensions to oblique-incidence inspections are also feasible. The SIN ratio alone cannot be used to fully assess the probability of flaw detection (POD). POD calculations require a complete understanding of both mean noise levels, and the manner in which the noise varies about its mean. Although we have made a good start toward understanding and quantifying the relationship between average and peak noise levels, more work is required on that front.

The above accomplishments may be summarized as follows:

• Developed experimental techniques and data-acquisition software for measuring noise levels and noise spectra.

• Developed three models for predicting absolute backscattered noise levels (and other noise characteristics) seen in a given UT inspection:

• ISMTB: For toneburst-pulse inspections. Predicts rms average noise level.
• ISMBB: For broadband-pulse inspections. Predicts rms average spectral components.
• Monte-Carlo Model: For arbitrary pulse types. Predicts typical noise wave-forms, and hence any average or peak noise characteristic.

The first two models relate backscattered ultrasonic noise to a frequency-dependent material figure-of-merit (FOM) and to details of the measurement system.

• Performed numerous tests of the models using noise data gathered from specimens with simple microstructures (Cu, Stainless Steel, Alpha-Ti) and from titanium alloy specimens supplied by engine manufacturers.

• Demonstrated (in collaboration with J. H. Rose) that the FOM could be predicted from first principles for single-phased, equi-axed, randomly-oriented microstructures.

• Measured and analyzed noise data from representative Ti-6246 specimens. Noted strong dependence of noise level on direction of sound propagation in some specimens.

• Used x-ray diffraction to investigate the sources of the noise anisotropy. It is now thought to arise from localized texture within the boundaries of "prior beta grains".

• Prepared specimens containing artificial hard-alpha material (oxygen contaminated case layer, nitrogen contaminated volumes). Measured sound speeds. Analyzed results using theories based on "rule of mixtures".

• Demonstrated how hard-alpha detectability could be assessed by combining noise models with models which predict echoes from defects.

• Predicted signal-to-noise ratios for focussed and planar transducer inspections of embedded hard-alpha inclusions of various diameters.

We have thus developed a firn scientific foundation for understanding the interrelationship of material and measurement parameters in determining the detectability of small flaws, particularly hard-alpha inclusions in titanium alloys. The next steps in the application of this knowledge should include: 1.) more extensive validation studies in collaboration with potential users; 2.) their use of the models to optimize the design of inspection systems for billets and other components where normal-incidence inspection is preferred; and 3.) use of the models to improve the quantification of detection reliability based on an analysis of field and laboratory data.