摘要

The basic theories of local decisions and data fusion techniques are depicted. The common flaws recognition in heavy cylinder ingots has been studied with data fusion techniques using ultrasonic method. Different flaws have different sensitivity to different types, positions and stances of probe, and data fusion techniques can make the full use of the information. Local decisions of each group of transducers can be made through perceptron algorithm according to the time and frequency domain characteristics of flaws. A data fusion based global decision is made in which the criterion is based on the maximum posterior probability of different types of flaws. The recognition results can be improved by using data fusion techniques in decision-order in which three groups of different types, positions and stances of transducers are used.