Bayesian inference is a statistical methodology wherein probabilities are calculated and modified based on “priors,” that is, the current state of knowledge. We utilize Bayesian inference methodology to assess the reliability of ultrasonic testing of weldments in accordance with the California Building Code and American Welding Society (AWS) D1.8 – Structural Welding Code-Seismic Supplement (a supplement to AWS D1.1 – Structural Welding Code-Steel). Utilizing reasonable estimates of the likelihood that a skilled welder will install a weld with a rejectable UT indication and the requirements for qualification to perform ultrasonic testing as specified in Annex F (Normative) of AWS D1.8, we show that a rejectable indication reported by a technician qualified at the minimum level of Annex F has approximately a 75% likelihood of being a Type 1 error. That is, under this set of assumptions (i.e., this set of priors), there is approximately a 75% probability that no rejectable defect exists at the location reported. We then analyze some of the factors that lead to this high probability of so-called “false calls” and the economic consequences of such reports. We briefly describe possible remedies to this situation.

Related Publications