In previous work by the present authors a model was developed for the estimation of the tensile properties (strength and ductility) of ferrous alloys intended for the welding of high strength low alloy steels used in the construction of ships. The method used to model the properties as a function of a large number of variables was based on a neural network within a Bayesian framework. This method is particularly useful when attempting to understand complex non-linear phenomena where the distribution of data within the input space is not obvious. In the present work, a similar approach is used to model the toughness (characterised by Charpy and dynamic tear tests) of the same alloys. The level of noise in the experimental data is perceived to be high, but it has nevertheless been possible to recognise reasonable trends and uncertainties when making predictions. For example, the toughness shows a non-linear deterioration as the oxygen concentration is increased; this behaviour is expected but can now be expressed quantitatively.
Read full abstract