Adaptive Learning Network Approach to Defect Characterization

The overall objective of this work was to demonstrate feasibility of adaptive nonlinear signal processing techniques applied to characterization of ultrasonic nondestructive testing (UNDT) waveforms for accurate inferences of flat -bottom hole sizes. The classified waveforms were ultrasonic pulse echoes obtained from two different sets of 7075-T6 aluminum area-amplitude test blocks and three different transducers. The eight flatbottom hole defect sizes ranged from 1/64 to 8/64 inch in steps of l/64 inch. Disciplines Materials Science and Engineering | Structures and Materials This 5. signal processing is available at Iowa State University Digital Repository: http://lib.dr.iastate.edu/cnde_yellowjackets_1975/7 ADAPTIVE LEARNING NETWORK APPROACH TO DEFECT CHARACTERIZATION Anthony N. Mucciardi Adaptronics, Inc. Mclean, Virginia The overall objecti ve of this work was to demonstrate feasibility of adaptive nonlinear signal process ing techniques applied to characterization of ultrasonic nondestructive testing (UNDT) waveforms for accurate inferences of flat -bottom hole sizes. The classified waveforms were ultrasonic pu l se echoes obtained from two different sets of 7075-T6 aluminum area-amplitude test blocks and three different transducers. The eight flatbottom hole defect sizes ranged from 1/64 to 8/64 inch in steps of l/64 inch. The ultrasonic equipment used in these studies was specially selected to provide as great a bandwidth as possible so that the ultrasonic waveforms would have maximum information content. Ultrasonic pulses were generated and directed into area-amplitude aluminum test blocks. The pulses reflected by circular f l at bottom holes were received by the transducer and amplified in the ultrasonic instrumentation. The pulses were recorded in a Biomation 8100 transient recorder. The Biomation recorder digitized the received signal and stored the digital information in its shift register memory. The pulse information was played back for inspection on an osci l loscope and transmitted to the memory of a Supernova computer for further processing. The model 8100 has a 2,048 word memory, so sampling at 100 MHz permitted storage of 20.48 ~sec of data. This 20.48 wi ndow was sufficient to cover the signal of in terest on these test blocks (i.e., its pul se reflected from the defect) if the start of the sampling window was delayed by 20 ~sec. A block diagram of the data acquisition system i s shown i n Fig . 1. The test specimens containing an artificial defect used in this project were area-amplitude test blocks prepared in accordance with ASTM El27-64. These blocks were fabricated from 7075-T6 aluminum alloy. Each of the test blocks is 3.75 inches long and 2 inches in diameter. Flat-bottomed holes are drilled along the axis of the block from one end to a depth of 3/4 inch. The holes vary in diameter from 1/64 to 8/64 inches in increments of l/64 inch. A typical bl ock is il l ustrated i n Fi g. 2. Two different sets of test blocks were fabricated by Trienco. The second set, made by Automation Industri es, was used to provide an independent set of blocks. In addition to the area-amplitude blocks, a special block was available to use as a reference standard. This block was 3 i nches long and contained no holes. The backwall echo was recorded before and after tests made wi th the other blocks and a comparison of the before and after tests was used to confirm that the operating conditions of the test system had not changed. A set of equipment was selected for this project as shown in the block diagram of Fi g. 1 . A Panametrics 5050PR pulser-receiver was used for generati ng the ultrsonic pulse, which drives the transducer, and for amplifying the