Artificial neural network predictors for mechanical properties of cold rolling products

Controlling product mechanical properties is an important stage in steel production lines. Conventionally, direct tensile tests are employed for this purpose: but their disadvantage is their high cost. The main objective of this paper is to develop an intelligent indirect method based on Artificial Neural Networks (ANN) for monitoring product mechanical properties without the need for expensive laboratory tests. The inputs into the proposed intelligent system include a wide variety of parameters from all production stages which it uses to predict such properties as Yield Strength (YS), Ultimate Tensile Strength (UTS), and Elongation (EL) as output. Moreover sensitivity analysis is performed based on using ANNs trained by data from three different grades because changing domains of input parameters is wider in these sets of data. Results show that the reduction in skin pass, the thickness after tandem and the ratio of Nitrogen to Aluminum are the effective parameters for all three mechanical properties among other inputs. Also, the thickness reduction in tandem affects the YS and EL values significantly, but UTS is not sensitive to this parameter noticeably. The variation of Vanadium content changes UTS value considerably. (C) 2011 Elsevier Ltd. All rights reserved.