The objective of this study was to develop a nondestructive test procedure for detection of fatigue cracks in steel anchor rods in auxiliary highway structures. Fatigue cracking in anchor rods generally occurs at locations that are difficult or impossible to inspect with visual inspection alone. Ultrasonic testing (UT) can be used for crack detection in anchor rods, but many state DOT’s and highway agencies do not incorporate UT in routine inspection programs due to the associated cost. Instead, UT is reserved for special inspections where a potential crack has first been identified through visual assessment. The method sought by the research would be implemented as a rapid screening test able to detect the presence of a fatigue crack with greater reliability than visual inspection.
The main focus of the research was an experimental evaluation of the proposed method for fatigue crack detection. A full-scale structure-foundation anchorage assembly was constructed for this purpose that included: eight 2 in dia. x 3 ft-9 in F1554 Gr36 steel anchor rods partially threaded at both ends with a 12 in threaded length and a thread pitch conforming to the ANSI/ASME B1.1 Unified Coarse Thread Series (4.5 threads/in); a 2 in thick A709 steel baseplate with 2-5/16 in dia. holes; an embedded 1/4 in thick A709 steel anchor plate; heavy hex nuts that measure 3.5 in (width across flats) x 2-13/64 in thick; and 4 in dia. x 1/4 in washers.
To perform a test, an accelerometer is coupled to the top surface of an anchor rod, and the surface is struck with an instrumented hammer. Evaluation of the results for the purpose of damage identification involves measuring the variation in the normalized response signal from a known baseline (healthy or uncracked condition). In order to identify indicators of fatigue cracking, several damage-sensitive features extracted from univariate and multivariate regression models were evaluated, including alpha-based regression coefficients, angle coefficients, cosh spectral distances, and regression residuals. Of the investigated features, alpha-based regression coefficients was found to be the most reliable indicator of anchor rod cracking. Since the recorded data is a collection of discrete points, regression models are employed to develop best-fit functions for statistical comparison. Autoregressive models (AR) are a particular classification of regression models that relate the current value of a predicted time series to past values of the same series. In this research, the AR models were used to fit accelerometer recordings normalized by the applied impulse.
From the experimental program it was found that by establishing a baseline measurement for an uncracked rod, and evaluating the change in Mahalanobis distance between the alpha coefficients of regression models fitting the test data, the test method was able to identify artificial cracks at the base of the leveling nut (a region known to be susceptible to fatigue cracking), extending 1/4 and 1/2 of the rod diameter in depth, with at least 95% confidence. Repeatability of the test approach was influenced by transducer-to-rod coupling conditions, and to a lesser degree, by the consistency of the mechanical impact.