This paper describes the work undertaken by the University of Glamorgan and CORUS Rotherham UK to apply artificial neural networks to model the cold alloy-steel bars and the heat treatment parameters with their end-product quality characteristics. Standard multi-layered feed forward artificial neural networks (ANNs) were employed to represent the functional mapping of inputs such as physical dimension, material composition and the parameters of the heat treatment cycles to the Brinell Hardness (HB) and the Ultimate Tensile Strength (UTS). The HB and UTS networks were validated with new data sets and demonstrated a satisfactory level of predictions over a range of conditions. These neural networks were then integrated into a Genetic Algorithm (GA) search strategy to identify the best material characteristics and furnace operating parameters in order that both the HB and UTS values are maximised. The results demonstrated that the hybrid strategy of combining the neural network based models with GA can deliver sensible results.

This content is only available via PDF.
You do not currently have access to this content.