Neural network application in aluminium hot-roll process

This paper reports the application of ANNs-based identification of the model of the aluminium hot roll process in the ALUMIX plant of Fusina, Venice (Italy). The neural model has been trained with the production reports of several months and, after validation, it has been embedded into a "job card calculator". Roughly speaking, this "calculator" implements an optimisation procedure, which looks for the minimum processing time with a limited rolling force. As a result, test rolls obtained with the proposed procedure have shown higher quality than the usual. In the paper, the plant is described, along with the neural model and its training procedure. This model is compared with several analytical models, showing the superior performance obtained through ANNs. Finally, the structure of a job card calculator and the experimental results are described.