Online mechanical property prediction and automatic product release for hot rolled steel strip products using artificial neural network models

Abstract

Tata Steel UK manufactures hot rolled strip products designed for a wide range of market applications. Typical mechanical properties for hot rolled strip products comprise: yield strength (Rp0.2), ultimate tensile strength (UTS) and elongation (A80). Physical tensile testing of a representative sample taken from each coil is generally used to measure these mechanical properties. As an alternative, artificial neural network (ANN) models have been developed to predict these tensile properties. This article describes the development and integration of ANN models for two hot rolled strip products: a low carbon steel for cold forming, DD11, and a high strength low alloy, S355MC. The current work proposes to combine the ANN model with a product release criteria to ensure high prediction confidence after deployment into the industrial environment. Following a period of 6-month implementation of the ANN models, predicted mechanical properties for both products were shown to be well within product specification requirements. The average predicted mechanical properties for DD11 were 241 MPa for Rp0.2, 351 MPa for UTS and 38.1% for A80. The average predicted mechanical properties for S355MC were 400 MPa, 488 MPa and 31.4% for Rp0.2, UTS and A80, respectively. Input parameters with relatively high Pearson correlation coefficient value showed that Rp0.2 increases monotonically as Mn content increases or coiling temperature decreases for DD11. Whereas for S355MC, Rp0.2 increases as Nb content increases, but coiling temperature did not show a linear relationship with Rp0.2.