ABSTRACT
Point kernel codes that simulate gamma-ray transport often use build-up factors to take scattered photons into account. This study introduces a new method, for computing multi-layer shield build-up factors. This method, based on an empirical formula for calculating double-layer shield build-up factors, is iterative. For an N-layer shield, each iteration of the method treats the first and the second layer of the shield. It replaces these layers by a single equivalent layer composed of an appropriate material and, hence, it turns the N-layer shield into an (N - 1)-layer shield. In order to determine the equivalent layer of an appropriate material, a neural network approach is developed: some neural networks trained on a large set of various configurations provide the equivalent material for any double-layer configuration. The method is implemented into MERCURE-6.3 straight-line attenuation code and is validated by comparison between MERCURE-6.3 results and reference data for one-dimensional geometries. Reference data obtained from transport calculations performed using the Sn transport code TWODANT. The comparisons prove the accuracy and sturdiness of the method.