Modelling imprecise and scattered multidimensional data using granular data compression and multiple granularity modelling

In this paper a systematic modelling approach is presented, involving two algorithmic procedures: a) a data pre-processing algorithm using granular computing and statistics and b) a granular neural-fuzzy ensemble network consisting of multiple granularity models. Both algorithmic procedures aim to reduce the data and modelling scatter often found in real industrial data. The study focuses on predicting the mechanical property of heat treated steel, in particular Charpy Toughness. This mechanical property yields high data scatter caused by unknown underlying fractural dynamics. The proposed methodology is shown to successfully model the process under investigation using a real industrial data set.