Evaluating the effect of error in thickness evaluation on the reliability of pipe remaining life prediction

Background Accurately estimating the remaining life of pipes affected by wall thinning relies on ultrasonic testing (UT). However, unpredictable measurement errors can complicate these estimations, and assuming normally-distributed errors may not always be justified. Purpose This study examines the impact of unpredicted UT measurement errors on pipe life estimation and explores suitable predictive models under realistic error assumptions. Methods Thirty artificially corroded plate samples were tested using conventional UT with a 5 MHz transducer, conducted by an unqualified operator. The distribution of measurement errors was assessed. Simulations modeled annual UT measurements for a virtual pipe over ten years, predicting thickness in the 10 th year using either linear regression or Bayesian inference. Results The analysis showed that the UT measurement errors did not follow a normal distribution, indicating limitations in using traditional statistical models. Simulations demonstrated that Bayesian inference consistently produced more reliable predictions of pipe thickness across different error models, outperforming conventional linear regression. Conclusion Assuming normality in UT error distribution is not appropriate in all cases. Bayesian inference offers a more robust and reliable approach for predicting the remaining life of pipes under uncertain error conditions compared to conventional methods.