Neural Networks www msm cam ac ukphasetrans There

Neural Networks www. msm. cam. ac. uk/phase-trans

There are many problems where simplification is unacceptable

Charpy fatigue tensile critical stress intensity corrosion

• Given a comprehensive description of material, process and structure, it is not yet possible to predict most properties.

Variables • • • C, Mn, Si, Ni, Cr, Mo, V, Co, B, N, O…. . Thermomechanical processing of steel Welding consumable Welding parameters Subsequent heat treatment

y x Michael Mc. Intyre

Solution • non-linear functions • large numbers of variables • uncertainties • exploit large knowledge base

Empirical Equations y = a + b (%C) +c (%Mn) + d (%Ni). .

y = a + b (%C) +c (%Mn) + d(%C x %Mn) y = sin (%C) + tanh (%Mn)

Hyperbolic Tangents

y A B x

What is the range over which an empirical method should be used?

Outliers

Cole & Bhadeshia, 1999

GTA weld at 823 K (data from Nippon Steel) 600 500 400 300 200 100 0 20000 30000 Life / hours 40000 Cole & Bhadeshia, 1999

Cool, 1996

Cool, 1996

As-welded 600 °C 700 °C 650 °C Cool, 1996

Siemens Mitsui Babcock Nippon Steel ABB

Murugananth & Bhadeshia, 2002

Coalesced bainite Keehan, Karlsson, Andrén and Bhadeshia, 2005

Nickel base alloy FT 750 dc wt%

Tancret & Bhadeshia, 2002 Yield stress / MPa 1000 800 600 400 200 0 0 200 400 600 800 Temperature / °C 1000 1200

International Fusion Reactor Reduced activation steels

Fusion Reactor Steels Kemp, Cottrell & Bhadeshia, 2006

Thank you

Classification networks