Low-alloy steel weldment creep life prediction using neural network analysis of ultrasonic backscattering

For welded components in the creep regime, premature creep cracking mostly occurs at the heat-affected zone (HAZ) and its adjacent areas in both base metal and weld. Therefore, methods for the detection of creep damage in wel dment; and practical procedures to relate the damage state to its remaining life are technically important. This experimental work was directed to develop a technique to predict life time of welded parts using creep damage information obtained from ultrasoni c testing. Welded 1Cr0.5Mo low-allow steel was used. Creep rupture tests we re conducted. Creep damage information was obtained from the ultrasonic bac kscattering at temperatures ranging from 600 and 680 degreesC and at stress es ranging from 70 to 120 MPa. From the creep curve, the minimum creep rate was determined. Ultimate tensile strength tests were also conducted on age ing specimens exposed together with the creep specimens during the creep te sts. The Fast Fourier Transformation (FFT) technique was used to obtain the power spectrum in each backscattered B-scan data. Correlation between minimum creep rate and experimental parameters/results (for both creep and ultrasonic testings) has been established based on an a rtificial neural network. The recurrent multi-layer perceptron (MLP) type o f neural network was selected in this study.