Hydrogen doesn't "form" after welding is complete. Atomic hydrogen can be introduced into the molten weld pool during the welding operation. Common sources of hydrogen include oil, grease, flux, paint, oxide products that contain hydroxides, etc. Under the influence of the high temperature arc, many of these products disassociate into the chemical components. Hydrogen easily goes into solution in the molten weld pool, however, upon cooling and solidification, the solubility is reduced. A good portion of the hydrogen comes out of solution and simply dissipates in the atmosphere. The remaining hydrogen can become trapped in what are known as hydrogen traps and within the atomic lattice of the solidified metal.

One hypothesis is that over time the hydrogen in solution diffuses through the atomic lattice and joins with other hydrogen atoms to form molecular hydrogen that exerts a stress on the surrounding lattice. If the lattice is brittle, it cracks. Over the course of several hours the crack propagates and eventually breaks the surface of the metal and is detected. Another hypothesis is the hydrogen combines with carbon to form methane gas. The molecule of methane (CH4) is much larger than molecular hydrogen (H2) and exerts much more strain on the surrounding lattice structure and again, a crack is formed.

Both hypothesis recognize the significance of martensite. Martensite is strong, hard, but brittle. In the first case the cracks form because the microstructure is hard and brittle. In the second case it is recognized that martensite is a body centered cubic crystalline structure that is supper saturated with carbon, thus the BCC must elongate to accommodate the excess carbon. The abundance of carbon makes it a plentiful source of carbon with combines easily with the hydrogen to form the methane.

It is said that the first hypothesis only accounts for 10% of the cracks observed. The second hypothesis is said to correlate with about 90% of the cracks observed. I leave it up to you to decide which hypothesis sounds more reasonable.

Both hypothesis recognize the time needed for diffusion of atomic hydrogen to take place. Thus, it may take several hours or more for the cracks to become apparent. High strength steels are more prone to delayed cracking, hydrogen assisted cracking, cold cracking (it is known by several names) because they have a higher carbon equivalency. The higher the carbon equivalency, the more likely the steel will develop a martensitic microstructure if the weld is allowed to cool too slowly. The more martensite present and the more hydrogen introduced into the weld pool, the more likely hydrogen assisted cracking will happen.

That's the short version.

Now, the hydrogen bake out (DPH) is carried out immediately after the weld is completed. It is best if the operation be completed as soon as possible before cracking has initiated. The weldment is heated to around 650 degrees F to "open the lattice" to allow the hydrogen to effuse into the atmosphere before it can cause damage, i.e., cracks. The weldment must be held at the high temperature for several hours to allow the hydrogen gas to diffuse through the lattice and find its way into the atmosphere. Other post weld heat treatment, i.e., thermal stress relief, normalizing, annealing, etc. accomplish the same results, but they are typically more expensive because of the higher temperatures required.

ASME required a stress relieve operation on steels that are more susceptible to delayed cold cracking, i.e., thick sections and those that are more highly alloyed. It accomplishes the same thing as the hydrogen bake out.

Al