The most simple, but more computationally intensive, is a transient coupled thermal-displacement step. Simply apply a predefined field as a temperature of 20degC and ramp your temperature boundary condition to -30degC over a specified time with a number of increments. Constrain other BC's in the usual way. The hard part may be determining the rate of temperature decrease (step/ramp/curve/etc) or the use of a temperature BC compared to a surface film condition.
I'm modeling something similar, which is thermal strain of a cam cover/seal/head assy. I'm using predefined fields as explained by MrMyers, but would like to have more information on the surface film condition, if I may.
Only use coupled temperature displacement models where the change in shape of the object will affect the thermal boundary conditions. Otherwise run the thermal analysis first, save the temperatures and then load them back into the stress model. Much quicker in run-time too.
Film conditions on surfaces will depend on various factors such as flow rate etc. If you have any real data then the best way is to 'fudge' the boundary conditions (to use a technical term) so that your model results will be approximately the same as your measured temperatures.