I think you need excitation too.
It is different sound when you shape noise with filter bank
and different when you have periodic excitation before filter bank. You could do this numericaly by using
inverse FFT and excitation (some kind of pulses, even basic aproximation of Dirac pulse with unit amplitude).
Excitation should be related to machine speed (like number of revolutions per second). Ford tried something like this to model engine noise in their cars.
Dubravko