Steel for fusion power plants - Phase

Materials for fusion power plants
Stéphane Forsik - Phase Transformations and
Complex Properties Group
www.msm.cam.ac.uk/phase-trans
FUSION POWER PLANT
The production of electricity from the fusion of deuterium and tritium is an alternative source to
classical fission nuclear power plants but, as a lot of technical issues still have to be resolved,
the first commercial fusion power plant is not expected before 50 years.
The plasma of deuterium and tritium will reach about 100 million K and neutrons with an energy
of 14 MeV produced during the reaction will strongly irradiate the materials constituting the first
wall of the reactor. Average irradiation doses are expected to be about 200 displacements-peratom (dpa) over a lifetime in service of several decades. The behaviour of materials in such a
hostile environment is not well-known and the choice of a candidate material with the best
mechanical resistance to radiation-induced hardening and swelling is crucial.
Large quantities of experimental data from irradiation experiments in fission reactors are
available. A neural network model (a regression method able to fit non-linear functions) was
trained on that database. It allowed us to catch complex relations between several input
parameters (chemical composition, heat treatment, irradiation parameters, etc.) and the output
(yield strength, DBTT, etc.) and helped in the understanding of some aspects of the irradiation
hardening. Such a model based on fission reaction data can also be extrapolated to higher
values in order to predict the behaviour of materials in a fusion environment.
ITER: prototype of a fusion reactor.
A model trained on a database containing 28 input parameters and about 1,600 lines was used to make predictions.
PREDICTION
EXTRAPOLATION
Neural-network predictions of the yield strength as a function of the
irradiated dose and tensile test temperature for two different alloys are
given and compared against experimental values:
The highest dose available in the database is 72 dpa whereas
~200 dpa will be reached in a fusion reactor. Two different
regression tools were used to extrapolate at such high
irradiation doses: a neural network and another method based
on Gaussian processes.
Modified 9Cr-1Mo ferritic steel
irradiated at 2.9 - 3 dpa:
predictions are in agreement
with experimental values and
uncertainties are small. The
irradiation-induced hardening
decreases at high tensile test
temperatures (annihilation of
radiation-induced defects).
The two predictions differ violently. The neural network predicts
0 MPa just below 100 dpa, accompanied with large
uncertainties whereas the Gaussian process predicts a stable
value between 60 and 200 dpa with a relatively high
confidence. This difference implies that hardening mechanisms
at high doses should be more deeply understood.
Low-activation ferriticmartensitic EUROFER’97 steel
irradiated at 2.5, 7.5 and 9 dpa:
predictions in agreement, the
yield strength increases with the
irradiation dose and saturates at
~10 dpa. Contrary to the
previous example, the radiationinduced hardening does not
disappear at high temperature,
due the difference in irradiation
temperature.
CONCLUSIONS AND FURTHER WORK
Two models were compared and predictions obtained with the neural network as well as with the Gaussian process are in
agreement with experimental values but differ when extrapolated and no experimental data are available at high irradiation doses.
Several aspects of the irradiation-induced hardening mechanism, such as the irradiation-induced dissolution of precipitates, need
to be investigated and understood. However, Gaussian processes appear to be have the same accuracy at low doses and their
training is less time-consuming.