Find yourself a microwave oven transformer (we used two with the secondaries in series), power it with a variac, improvise an adjustable spark gap, and build a showering arc noise generator. For lower frequencies typical of conducted EMI, couple the noise into the power line with a 1:1 transformer consisting of a few turns in the primary and the secondary and a higher permeability toroidal core. You can even wind it with two secondaries for injecting common mode noise or wind a dozen or so turns around an air core for higher frequency radiated noise. Radiated noise immunity wasn't much of a problem for us though.
It won't be precise but the noise generated is broadband and it can be used to figure out what sorts of filtering is more effective for a given application. IIRC (it has been a while), an 1/8" gap was putting some pretty nasty 4KV crap on the line at frequencies up to around 10MHz with a 120V input. It would routinely lock up and/or fry just about anything we tested that didn't include some pretty robust filtering. I think we originally set our standard at 45V input with a 1/16" gap for about 1.5KV of noise.
We no longer use it though. Design engineering wanted something more consistent and insisted on our buying an actual pulse generator about 10 years ago. It will pass things set at 2.5KV that the showering arc noise generator routinely crashed at 1.5KV.