Increasing Exploitation Durability of Two-Layer Cast Mill Rolls and Assessment of the Applicability of the XGBoost Machine Learning Method to Manage Their Quality.

Abstract

The increase in exploitation durability of two-layer cast rolls with the working layer made of high chromium cast iron allows one to significantly improve the quality of rolled metal as well as to increase the economic efficiency of the manufacturing process. However, it is severely hindered due to the massiveness of castings, the impossibility of both evaluating mechanical properties along the depth of the working layer, and providing the structural uniformity of the working surface and the decrease in stresses. In our research, aiming to enhance the exploitation durability of sheet rolls, it is recommended to achieve structural uniformity by CuMg alloying, which increases the concentration of copper up to 2.78 wt.% in certain zones and, owing to the accelerated austenite decomposition at a high temperature during the cool-down of the castings, led to the reduction in excessive strength and the level of heat stresses in the castings. We propose the regimes of cyclic heat treatments which, due to the decomposition of retained austenite and the fragmentation of structure, control the level of hardness to reduce and uniformize the level of stresses along the length of a barrel. A further improvement in the predictions of exploitation durability using XGboost method, which was performed based on the chemical composition of the working layer of high-chromium cast iron and heat treatment parameters, requires taking into account the factors characterizing exploitation conditions of specific rolling mills and the transformations of structural-phase state of the surface obtained by a non-destructive control method. As the controlled parameter, the hardness measured on the roll's surface is recommended, while the gradient change in mechanical properties along the working layer depth can be feasibly analyzed by a magnetic method of coercive force measuring.