Theoretical design of ferritic creep resistant steels using neural network, kinetic, and thermodynamic models

Creep resistant ferritic steels based on Fe–2·25Cr–1Mo and Fe–(9–12)Cr have been used extensively in the energy and petrochemical industries for more than half a century. This is because they have an outstanding record of reliability in quite aggressive conditions over time periods as long as 30 years. The alloys have undergone progressive development throughout their history,1,2 particularly in order to permit the use of higher steam temperatures in power plant which can then operate with greater efficiency. The basic principles of alloy design for creep resistance are therefore well established and well founded on experience. The steels must have a stable microstructure which contains fine alloy carbides which resist the motion of dislocations; however, changes are inevitable over the long service time so that there must be sufficient solid solution strengthening to ensure long term creep deformation. There may be other requirements such as weldability, and corrosion and oxidation resistance. It is nevertheless difficult to express the design process quantitatively given the large number of interacting variables. It was the purpose of the present work to encrypt some of the published experimental data into a quantitative procedure for estimating the creep rupture stress of power plant steels as a function of the chemical composition, heat treatment, temperature, and time. The rupture stress is chosen as the variable to model because of the ready availability of data in the open literature, and because it is a very common parameter used in industry during alloy development. The model which is developed is then tested against known metallurgical trends and used to propose two new alloys which ought to have better properties than anything developed to date. We begin with a brief description of the neural network method used to develop the quantitative model. The method used here is due to MacKay;3–6 and it has been recently reviewed.7