Adaptive numerical methods, such as neural networks, have received considerable attention in recent years in relation to the modelling of complex physical systems. In this work a variety of such methods have been applied to the modelling/data mining of commercial materials production data, thereby avoiding the scale-up problems associated with laboratory scale investigations of materials behaviour. It is shown that adaptive numerical methods may determine valuable empirical models from such complex databases, whilst the value of transparent modelling methods (where the underlying relationships between input variables and modelled characteristics may be clearly visualised) is highlighted in providing model confidence and the potential to extract novel physical understanding.