Metamodel-based Markov-Chain-Monte-Carlo parameter inversion applied in eddy current flaw characterization

Abstract

Flaw characterization in eddy current testing usually requires to solve a non-linear inverse problem. Due to high computational cost, Markov Chain Monte Carlo (MCMC) methods are hardly employed since often needing many forward evaluations. However, they have good potential in dealing with complicated forward models and they do not reduce to only providing the parameters sought. Here, we introduce a computationally-cheap surrogate forward model into a MCMC algorithm for eddy current flaw characterization. Due to the use of a database trained off-line, we benefit from the MCMC algorithm for getting more information and we do not suffer from the computational burden. Numerous experiments are carried out to validate the approach. The results include not only the estimated parameters, but also standard deviations, marginal densities and correlation coefficients between two parameters of interest.

Exploring high-frequency eddy-current testing for sub-aperture defect characterisation using parametric-manifold mapping

“…The small dimensions of the coil are expected to further reduce the depth of penetration of eddy-currents within the test material [33]. While many authors have explored characterisation at lower frequencies, matching the skin depth to the defects of interest [7,19,22,16], here we examine the characterisation limits at higher frequencies with the aim of evaluating the uncertainty in characterisation of sub-aperture defect dimensions.…”

“…[13] have developed an approximate impedance integral model to characterise grain microtexture in Titanium superalloys [14] and use it to improve sub-mm defect characterisation in these materials [15]. Others have developed meta-models of the defect database and optimisation techniques to improve the speed of characterisation [16,17,18], while approaches employing artificial intelligence and machine learning methods have also been explore with some success [9,19,20,21,22].…”

New Proposal for Inverse Algorithm Enhancing Noise Robust Eddy-Current Non-Destructive Evaluation

“…A NN employed for a crack depth estimation uses back-propagation algorithm with forward propagation of input data, backward propagation of error, and consequent changes in input neuron weight values. The real measured data for each defect with known geometry are applied as the training sets and the applied training function is Bayesian regularization based on Levenberg–Marquardt optimization, [ 3 , 9 , 10 ]. The neural network consists of an input, a hidden and an output layer.…”

“…Various mathematical procedures are being sought and used with the aim of partial improvements of preciseness and reliability. Scientific works are focused mainly on reducing uncertainty in estimating parameters of defect geometry, especially by applying adaptive Monte Carlo method [ 2 ], metamodels-based Markov–Chain–Monte-Carlo [ 3 ], Genetic Algorithm [ 4 ], Neural Networks [ 5 ], Particle Swarn Optimization [ 6 ], and others [ 7 , 8 ]. Many published papers address the effect of a defect structure on response signals and then optimize 3D defects [ 9 ].…”

Exploring high-frequency eddy-current testing for sub-aperture defect characterisation using parametric-manifold mapping

“…Non-iterative approaches compute a database of possible solutions off-line for a given inspection geometry and defect parameter space, then use various techniques and algorithms to search and interpolate the database to find the closest matching results to experimental measurements. Authors have approached this from various directions in eddy-current including methods based in; machine learning techniques [9,10,11,12,13] and meta-model generation [14,15,16]. In spite of the impressive characterisation ability of these approaches to large defects, there remains a fundamental limit to the ability of these techniques to accurately size small surface-breaking defects, where sizing is of critical importance.…”