Improving pulse eddy current and ultrasonic testing stress measurement accuracy using neural network data fusion

Abstract

Stress and residual stress are two crucial factors which play important roles in mechanical performance of materials, including fatigue and creep, hence measuring them is highly in demand. Pulse eddy current (PEC) and ultrasonic testing (UT) are two non-destructive tests (NDT) which are nominated to measure stresses and residual stresses by numerous scholars. However, both techniques suffer from lack of accuracy and reliability. One technique to tackle these challenges is data fusion, which has numerous approaches. This study introduces a promising one called neural network data fusion, which shows effective performance. First, stresses are simulated in an aluminium alloy 2024 specimen and then PEC and UT signals related to stresses are acquired and processed. Afterward, useful information obtained is fused using artificial neural network procedure and stresses are estimated by fused data. Finally, the accuracy of fused data are compared with PEC and UT information and results show the capability of neural network data fusion to improve stress measurement accuracy.