Abstract
The present paper describes work carried out on a collaborative European Coal and Steel Community (ECSC) project, completed in 2000, which aimed to develop a quality prediction forecasting method for both the internal and the surface quality of the as-cast semi. The techniques used included mathematical models, artificial neural networks (ANNs) such as multilayer perceptron (MLP) nets and self-organising maps (SOMs) and other databased methods using 'fuzzy logic' and statistical techniques. Plant data were obtained for both carbon steel and stainless steel slab casting and also for billet casting, and were provided to the various partners to evaluate the above techniques. The various analyses are described in the present paper. The general conclusions are that training the MLP net was difficult owing to the lack of poor quality casting data and the SOM was satisfactory for monitoring general process behaviour rather than identifying the location of individual defects. A quality prediction system derived from combinations of these techniques was trialled successfully on plant.