Data-fusion method for flaws detection based on reliability evaluation

In order to solve the problem of low accuracy of flaws detection in non-destructive testing, a data-fusion method based on RBF neural network and evidential reasoning (ER) rule is studied. Firstly, a data-fusion model is proposed which contains two parts. Secondly, the first part RBF neural network is used to fuse the feature layer, and its output is normalized as the basic probability assignment (BPA) of each evidence for the next part. Then, according to the traditional D-S evidence theory, the accuracy of the flaws detection is not high enough and the evidence conflict is obvious. So it's necessary to introduce the evidence reliability factor to evaluate the reliability of the evidence, so that all kinds of evidence pass evaluation and then make fusion. Finally, the defect data of a composite material is obtained by means of ultrasonic detection, and the method proposed in this paper is verified. The results show that the method can accurately identify the flaw types, compared with other methods to improve the accuracy of detection.