Simulation Based Validation of the Detection Capacity of an Ultrasonic Inspection Procedure

Artikel i vetenskaplig tidskrift, 2012

With aging engines and structures, reliable maintenance will be a very important challenge in the future. This and other safety aspects have encouraged the development of non-destructive testing (NDT) technologies to detect in-service defects at an early stage. In order to quantify the inspection reliability the methodology of probability of detection (POD) was developed by the aeronautical industry in the early 1980s. This statistical tool reduces the number of artificially produced artefacts that needs to be introduced into the test blocks in order to get statistically valid information of the detection capacity, even though its legitimacy is limited to very restricted conditions. Even though this procedure reduces the number of defect it includes a large number of inspections and personnel and these campaigns thus tend to be time consuming and very expensive. This paper address the development of a procedure for generating POD based on synthetic data using NDT simulation software. The intention is to have an optimized experimental phase combined with much more efficiently retrieved simulated data. This has been achieved by fitting a multi-parameter prediction model to ultrasonic simulation software (simSUNDT) in an orthogonal design of experiments. The probabilities of detection as function of defect size (POD curves) at different defect depths were generated by Monte Carlo simulation introducing variations in the control factors with a physical interpretation in the emulator. The validation of this developed methodology has been based on a qualified ultrasonic procedure, UT-01. It specifies manual ultrasonic inspection of piping components within Swedish nuclear power plants and qualification of personnel according to this procedure has been in place since 1996. Experiences made by these qualifications were comprised as POD curves for both fatigue cracks and stress corrosion cracks in 2005. These POD curves are in this paper compared with corresponding generated with the above described simulation based methodology and the agreement found was remarkable well. © 2011 Elsevier Ltd. All rights reserved.