I think you're on the right track. If you're doing this in a spreadsheet, MIN, MAX, and probably IF are going to be your friends for enforcing cutoffs and boundaries (like non-negative damage, auto hits/misses, etc.).
I'm guessing what you're trying to get a feel for is not (just) the average expected damage per round, but the distribution. That might be more challenging to do in a spreadsheet (although totally possible). My Spreadsheet Fu is lacking, so others will likely be able to provide more guidance as to specific strategies.
That being said, I threw together a [EDIT: now somewhat less] simple interactive widget using Jupyter with dyce¹ that might help. Assuming I understand your mechanic (and got my math and my widgets wired up right), the amount of damage a PC can expect to land on a single round with a TH mod of -2, a DC target of 11, 2d6 on attack, 1d6 on defense, a DMG of 3, Armor of 1, and an RoF of 6 is:
This models a TH and DC that are both determined by d20 roll each round (which I assumed because pools of d6s were eligible to be put in play for both attack and defense). If that's wrong (i.e., only the PC rolls, and it's against a static DC if the PC is attacking or a static TH if the PC is defending), it shouldn't be too difficult to correct for. EDIT: This now models three scenarios²: a PC attacking (supporting crit hits/misses against the NPC); a PC defending (supporting crit hits/misses against the PC); and PC vs. PC (supporting symmetrically-negating crit hits/misses). The code is a little unwieldy, but the critical functions are expected_dmg_frm_rnd_pc_attacks, expected_dmg_frm_rnd_pc_defends, and expected_dmg_frm_rnd_pc_v_pc.
anydyce³ is used to to generate "burst" graphs. You can play around with it in Binder: 
[source GitHub Gist]
A new instance may take awhile to launch if that link hasn't been followed in awhile. Binder will delete a launched instance after a period of inactivity, so download any work you want to save.
I'm hoping that even if you don't speak Python, the above is accessible enough to either get you where you want to go calc-wise, or give you enough inspiration to modify your spreadsheet to get it to do what you want. Like I said, I think you're close.
¹ dyce is my Python dice probability library.
² The prior versions of the gist and binder are available, if helpful.
³ anydyce is my visualization layer for dyce meant as a rough stand-in for AnyDice.