Efficient Global Sensitivity Analysis of Model-Based Ultrasonic Nondestructive Testing Systems Using Machine Learning and Sobol’ Indices

Abstract

The objective of this work is to reduce the cost of performing model-based sensitivity analysis for ultrasonic nondestructive testing systems by replacing the accurate physics-based model with machine learning (ML) algorithms and quickly compute Sobol' indices. The ML algorithms considered in this work are neural networks (NN), convolutional NN (CNN), and deep Gaussian processes (DGP). The performance of these algorithms is measured by the root mean squared error on a fixed number of testing points and by the number of high-fidelity samples required to reach a target accuracy. The algorithms are compared on three ultrasonic testing benchmark cases with three uncertainty parameters, namely, spherically-void defect under a focused and a planar transducer and spherical-inclusion defect under a focused transducer. The results show that NN required 35, 100, and 35 samples for the three cases, respectively. CNN required 35, 100, and 56, respectively, while DGP required 84, 84, and 56, respectively.

Magnetic characterization of steel strips using transient field measurements: global sensitivity analysis and regression from a machine-learning perspective

“…Furthermore, the less accurately estimated JA parameters agree with the less sensitive ones, which is to be expected. The preliminary use of SA techniques proves useful in order to select a subset of the most important variables for inversion purposes [22]. The perspectives of the presented work are many.…”

“…Once the database is created, a metamodel is built from it using a Gaussian process regressor (GPR) [22,23]. One can define the approximate solution of the target as M : X → Y based on the training set D train = {(x 1 , y 1 ) , (x 2 , y 2 ) , .…”