Structural health monitoring ultrasonic thickness measurement accuracy and reliability of various time-of-flight calculation methods

Abstract

Abstract. The accuracy, precision, and reliability of ultrasonic thickness structural health monitoring systems are discussed in-cluding the influence of systematic and environmental factors. To quantify some of these factors, a compression wave ultrasonic thickness structural health monitoring experiment is conducted on a flat calibration block at ambient temperature with forty four thin-film sol-gel transducers and various time-of-flight thickness calculation methods. As an initial calibration, the voltage response signals from each sensor are used to determine the common material velocity as well as the signal offset unique to each calculation method. Next, the measurement precision of the thickness error of each method is determined with a proposed weighted censored relative maximum likelihood analysis technique incorporating the propagation of asymmetric measurement uncertainty. The results are presented as upper and lower confidence limits analogous to the a 90/95 terminology used in industry recognized Probability-of-Detection assessments. Future work is proposed to apply the statistical analysis technique to quantify measurement precision of various thickness calculation methods under different environmental conditions such as high temperature, rough back-wall surface, and system degradation with an intended application to monitor naphthenic acid corrosion in oil refineries.

Ultrasonic thickness structural health monitoring of steel pipe for internal corrosion

“…The thickness error t e can be positive or negative defined as the difference between measured thickness t m and true thickness t t as shown in Equation 8.3. Figure previously published [Eason et al (2016a)].…”

“…A confidence region of the most likely µ and σ parameters corresponding to the relative likelihood values greater than α as derived from a χ 2 distribution with two degrees of freedom is shown in Equation 8.14 [Meeker and Escobar (1998)]. The industry recognized a 90 confidence value can be determined as the cumulative distribution point from the maximum likelihoodμ andσ model, while the a 90/95 confidence value can be determined with the Delta method to establish Wald confidence intervals [Annis (2009)]; alternatively, the a 90/95 confidence value can be determined using a simulation method similar to Monte Carlo to construct a set of distribution models with the µ and σ parameters from the relative likelihood confidence region perimeter [Eason et al (2015b[Eason et al ( , 2016a].…”

“…The relative likelihood method does not capture an individual measurement data point confidence interval, and the censored relative likelihood method does not consider if an individual measurement data point mean has asymmetric uncertainty. However, the individual data point mean and asymmetric measurement confidence intervals are considered in the proposed weighted censored relative likelihood method [Eason et al (2016a)] as described in Equations 8.16 or 8.17 with κ as the weight parameter ranging from κ = 0 for no censoring (same as Equation 8.12) to κ = 1 for full censoring (same as Equation 8.15).…”

“…An experiment and analysis was completed using sol-gel transducers in pulse-echo mode on a calibration block with a flat back-wall to demonstrate the weighted censored relative likelihood method for various thickness calculation algorithms [Eason et al (2016a)].…”

“….2 AISI 1018 carbon steel eleven step calibration block from 24.00 mm to 25.00 mm with 44 direct sol-gel transducers. Figure previously published[Eason et al (2016a)]. 8.3.1.2 Time-of-Flight Calculation MethodsWhile many different thickness calculation methods exist[Svilainis (2013)], this work addresses relatively simple time-of-flight methods; the initial focus is not to discover the most precise method, but rather, avoid the least precise among commonly used methods.…”