NDT image fusion using eddy current and ultrasonic data

A data fusion approach to the classification of eddy current and ultrasonic measurements is proposed in a context of defect detection/recognition methods for non‐destructive testing/evaluation systems: the purpose is to demonstrate that a multi‐sensor approach that combines the advantages carried by each sensor is able to locate potential cracks on the inspected specimen. Different approaches have been compared: a pixel level data fusion approach, that distinguishes between the defect area and the no‐defect areas, by means of the information carried by the intensity of each pixel of the eddy current and ultrasonic data; a feature level data fusion approach that uses the features computed on the measured data; a symbol level data fusion approach that extracts symbols from the two sensors as complementary information and classifies the data by using these symbols. The experimental results, carried out on an aluminium plate, pointed out the ability of the symbol level proposed approach to classify the input images within a minimum overall error, by taking into account the probability of detection and the probability of false alarm for the defect.