Session: Advanced Modelling for NDE

Paper Number: 170256

170256 - On Incorporating Realism for Finite Element Modelling With Applications to Surface Breaking Crack Characterisation 

Abstract: 

Non-Destructive Evaluation (NDE) is used in the power and aerospace sectors to detect and characterise defects in safety-critical components. In the field, practitioners commonly use Ultrasound Testing (UT) methods (e.g. pulse-echo, pitch-catch, time of flight diffraction) to detect, size and characterise defects such as surface-breaking cracks. Machine Learning (ML) could help practitioners to reliably analyse ultrasonic data obtained from pulse-echo inspections, as ML is increasingly well-suited to data interpretation with the growth in computing power and parallelisation.

The implementation of ML algorithms requires sufficient good-quality data to generate adequate training, validation and testing datasets, with the requisite labelling. Real defect data is difficult to obtain, and costly to acquire, and often only directly relevant to the specific inspection parameters used during collection. To overcome these limitations, this work utilises high-fidelity finite element modelling to generate practically realistic A-scans (time series) for ML algorithms. Our simulations incorporate accurate transducer characteristics and crack geometries (obtained from a collection of thermally fatigued samples loaned by industrial partners), while including improved procedural methods to mimic practitioner methodology. For example, scanning from both sides and manipulating the transducer to maximise amplitude values. The simulations are used by an ML algorithm to predict the tilt angle and height of a two-dimensional surface-breaking rough crack using pulse-echo UT.  Performance also remains relatively consistent when evaluating rough versus smooth defects.

Presenting Author: James Gaffney University of Liverpool

Presenting Author Biography: James works in the mathematics department at the University of Liverpool in the UK where he studies ultrasound for NDE.

Authors:

James Gaffney University of Liverpool
Stewart Haslinger University of Liverpool
Thomas Beckingham University of Liverpool
Daniel Colquitt University of Liverpool
Will Daniels Kande
William Christians Universtiy of Liverpool
George Sarris Imperial College London