Thermal cycling is a generalized term for adjusting (usually shrinking) grain size in steel after normalizing. Even if you are dealing with annealed steel delivered in a state from the retailer such that you can just heat and quench the steel, you may wish thermally cycle steel. At the end of this is a reminder of the progression of steps. You can start at any step if the material allows, including just heating and quenching bar stock steel without anything else in some cases, but you shouldn't really skip steps if you have to start at forging, for example, to shape a tool, and think you'd like to just go to quenching at the right heat just after hammering.
The process of thermal cycling is generally low temperature heats that are just to non-magnetic, or heats that start similar to a normalizing heat but then descend at each iteration. The intention of these thermal treatments is the same, regardless of which of those is true. By heating steel to a relatively low temperature, but one high enough to establish new grains, you are replacing the grain that is present with smaller grain with each successive heat. You will be able to see the effect of this in snapped samples, especially if you intentionally or unintentionally overheat a sample and need to correct the resulting enlarged grain.
Thermal cycling is sometimes referred to as normalizing in blacksmith circles, as in, all heats after forging and before quench may be referred to as "normalizing heats". I think this is confusing because normalizing is generally more effective if it's done at a temperature high enough to get material in solution.
If you heat steel below the point that it's nonmagnetic, as some cycles prescribe, the lower subcritical heats are really affecting carbides more than they are establishing any new grain. That is, time held subcritical (still magnetic, but relatively hot, like 1300ºF) will cause carbides to become more spheroid in shape. Actual spheroidizing in a controlled process is long in duration, though, so it's my opinion that this may not achieve much for our purposes.
My preference is to normalize, quench off of the last normalizing heat when the temperature is approaching nonmagnetic in color (magnetism may not return at this point after normalizing, but disregard that other than to know you you may observe steels not returning to magnetic even at black heat after a high temperature normalizing cycle), and then perform a minimum of three (usually four or five) thermal cycles after the quench.
If you don't quench after you normalize, and you may wish to avoid doing that - you can just reheat steel after it's become magnetic again.
I don't generally differentiate temperature -wise, but rather these thermal cycles are heating until just at the point that the steel is transitioning to nonmagnetic and then I allow the steel to rest until it would be close to black in a lit room. For chisels and plane irons, this doesn't take long - the total process after normalizing is only minutes.
Why quench after normalizing? it seems pointless because you're going to reheat the steel, but I see better results in some steels - minimally better, but better - especially in terms of hardness in something like 52100. I don't know why this is. There is a separate benefit, though, and that is when we quench steel, we bring it to a martensitic structure. Larrin Thomas addressed this at some point in the past - starting from a martensitic structure before thermal cycles makes them a little bit more effective.
The other part of this that I'm laying out here is that I see much better straightness of brine quenched chisels if they receive a pre-quench at the end of normalizing. Aside from the factual nature of thermal cycles being more effective with a martensitic (what we consider the state of steel used in our actual tools - after quench), I realize this is bordering on "pet process" type stuff. You should do whatever you see good results with and not what I do.
Too, if you watch something like forged in fire or talk to a metallurgist online, you may encounter people with preconceived notions that don't match your experience. Forged in fire's humorous advice comes out as "never put hot steel in water" or "you're just going to get cracks". if you're dealing with steel that can come up short hardening in oil, brine is the next reasonable step. if you're worried that it's too fast, you can warm the brine, but in dealing with W1 and other intentionally low hardenability steels - or 52100 in chisels where retained austenite can be a problem, the fast quench is needed. I've not cracked a chisel in brine. For stanley type plane irons, the cross section is thin - and knives, if you're thinking of making thin slicing kitchen knives - and that thin section will transition from hot to cold well enough in a fast oil. But water hardening chisels seem to like the brine and the result is a very crisp tool.
I've spilled my beans here, but I'll abandon exactly what I do the second something that works better reveals itself. Some time a couple of years ago, I sent samples of 26c3 to be tested, and I was confident in doing so as just heating the bar stock and making a chisel resulted in a good chisel. Adding thermal cycles and ensuring a quench strong enough to get full hardness made chisels that are as good as anything I've ever used, including the best japanese chisels. the samples that I sent averaged 63.8 hardness, and required twice as much energy on average to break in a side strike as O1 steel that is properly done and two points harder. if I get those results, you'll get those results if you do the same thing, and maybe better. those test results, which were on little coupons that may not be quite as good as you or I get with a larger piece of steel and more control over temperature - the coupons heat quickly, sometimes too fast......
...anyway, if we get those results or better heat treating by hand and eye, we're well beyond what you'll find in every aspect comparing something like V11 (harder, twice as tough, and far better edge stability due to presence of iron carbides instead of chromium), or O1 (harder, twice as tough), or A2 (as tough or tougher at high hardness and potentially higher hardness with better edge stability).
You see where this is going - results as good or better than you can buy or pay to have done for you. Thermal cycles and establishing new grains that are finer than those that come out of normalizing or that arrive in annealed stock, key.
I don't have a furnace - one of the reasons I don't is because these cycles would take forever in one, but you may have a large batch, and the waiting could be a non-factor for you. If you do have a furnace and you feel like the schedule is too long, you always have the option of running the thermal cycles in the open atmosphere and then pre-heating your furnace so that you can toss the cycled steel in the furnace before quenching.
So, the full process:
* forging
* normalizing (if not forging, but wanting to reset the grain and elements in a bar of steel)
* thermal cycling
* austenitizing
* quenching
With martensitic steel, you'll always do at least the last two.
There's a one-off here, and that is that there are some annealing processes that people will favor over thermal cycling. I don't do them, but you may wish to. You can read about imitating a DET anneal from Larrin Thomas if you travel to his site. One of the nice things about the site, even though it's a little intense for someone who is just trying to get some experience under their belt, is that Larrin will test a process and provide the results. At least in one case (80crv2) Larrin's toughness and hardness results didn't reveal any benefit to thermal cycling if steel went through a DET type annealing process.
For our purposes without having a furnace or wanting to prescribe it, some approximation can be done by taking steel that is just changing to nonmagnetic and putting it in a bucket of fine vermiculite to cool as slow as possible. If an item is thin and still cools to fast, you do this with a sandwich. I think thermal cycles are easier, and you never reach into a bucket to grab something you forgot was 2 hours in the bucket and not 2 days, and burn your fingerprints off.