Decision-level multi-method fusion of spatially scattered data from nondestructive inspection of ferromagnetic parts

TL;DR

This paper introduces a novel density-based fusion method for combining scattered defect detection data from multiple nondestructive testing sensors, effectively reducing false alarms and improving defect localization accuracy.

Contribution

It presents a new fusion technique that explicitly handles localization uncertainties and registration errors in multi-sensor nondestructive testing data.

Findings

Successfully reduces false alarms in defect detection

Enhances detection of small defects

Demonstrates robustness with experimental data

Abstract

This article deals with the fusion of flaw detections from multi-sensor nondestructive materials testing. Because each testing method makes use of different physical effects for defect localization, a multi-method approach is promising to effectively distinguish the many false alarms from actual material defects. To this end, we propose a new fusion technique for scattered two- or three-dimensional location data. Using a density-based approach, the proposed method is able to explicitly address the localization uncertainties such as registration errors. We provide guidelines on how to set all key parameters and demonstrate the technique's robustness. Finally, we apply our fusion approach to experimental data and demonstrate its ability to find small defects by substantially reducing false alarms under conditions where no single-sensor method is adequate.