Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[sGdYy409L]

1), 5) - To my knowledge, there is no way of changing quality parameters of items in arena mode (If someone knows otherwise, please chime in). As much as I would love to perform a study on the effects of quality level on crossbow performance, this seems to be an impossible venture - at least on a scale which would produce reliable results. I could train up 10 Marksdwarfs and try to set up a small scale experiment in Fortress mode, but this simply wouldn't be enough to be able to perform even basic statistical analysis and achieve meaningful results.

It is possible with dfhacks "changeitem" command. For changing items equipped by a creature use it like this:
- select the creature with v
- select a worn item with +/-
- enter (manually or via hotkey) the command "changeitem q 5". This changes it to be of masterwork quality.

[Pirate Bob]

1), 5) - To my knowledge, there is no way of changing quality parameters of items in arena mode (If someone knows otherwise, please chime in). As much as I would love to perform a study on the effects of quality level on crossbow performance, this seems to be an impossible venture - at least on a scale which would produce reliable results. I could train up 10 Marksdwarfs and try to set up a small scale experiment in Fortress mode, but this simply wouldn't be enough to be able to perform even basic statistical analysis and achieve meaningful results.

It is possible with dfhacks "changeitem" command. For changing items equipped by a creature use it like this:
- select the creature with v
- select a worn item with +/-
- enter (manually or via hotkey) the command "changeitem q 5". This changes it to be of masterwork quality.

I'll have to think about whether I can automate this along with what I am already doing.  I know you can use lua scripting with DFhack, so theoretically it should be possible.  We can definitely use this at least on a small scale to see if quality matters at all.

I am currently running bronze armor vs copper bolts with increasing IMPACT_FRACTURE.  I don't have time to post full results now, but (as I think Urist already knows) only steel ever produces bruises for contact area 2 or 10.  For contact area 10, the deflection cutoff for steel is 48 and the bruising cutoff is 116.  For contact area 2 defection cutoff = 42 and bruising 57.  Even adamantine never produces bruising.  I will be busy the rest of the day, so I may not be able to post again until tomorrow evening.

[Pirate Bob]

1), 5) - To my knowledge, there is no way of changing quality parameters of items in arena mode (If someone knows otherwise, please chime in). As much as I would love to perform a study on the effects of quality level on crossbow performance, this seems to be an impossible venture - at least on a scale which would produce reliable results. I could train up 10 Marksdwarfs and try to set up a small scale experiment in Fortress mode, but this simply wouldn't be enough to be able to perform even basic statistical analysis and achieve meaningful results.

It is possible with dfhacks "changeitem" command. For changing items equipped by a creature use it like this:
- select the creature with v
- select a worn item with +/-
- enter (manually or via hotkey) the command "changeitem q 5". This changes it to be of masterwork quality.

I'll have to think about whether I can automate this along with what I am already doing.  I know you can use lua scripting with DFhack, so theoretically it should be possible.  We can definitely use this at least on a small scale to see if quality matters at all.

I am currently running bronze armor vs copper bolts with increasing IMPACT_FRACTURE.  I don't have time to post full results now, but (as I think Urist already knows) only steel ever produces bruises for contact area 2 or 10.  For contact area 10, the deflection cutoff for steel is 48 and the bruising cutoff is 116.  For contact area 2 defection cutoff = 42 and bruising 57.  Even adamantine never produces bruising.  I will be busy the rest of the day, so I may not be able to post again until tomorrow evening.

[sGdYy409L]

I just noticed that the dfhack "changeitem" command can also be used to create proper bone bolts.
type:

changeitem m CREATURE:DWARF:BONE force

and enjoy your new dwarf bone bolts!

[Pirate Bob]

I have run for several values of IMPACT_FRACTURE for both copper bolts  vs. bronze armor and copper bolts vs iron armor, with all other params vanilla (including contact area=2).  Here are plots of the forces required for 50% deflection as a function of IMPACT_FRACTURE:

Spoiler (click to show/hide)

As you can see, deflection levels off abruptly at Force=69 for Bronze and Force=118 for iron.  I think this is because deflection abruptly drops to zero at this force:

Spoiler (click to show/hide)

This is for bronze.  Notice that for IMPACT_FRACTURE x 3, the deflection starts to go down smoothly, and then drops sharply to zero, while bruising continues to go down smoothly.  The cutoff in deflection may be the cutoff proportional to 1/IMPACT_STRAIN_AT_YIELD, but I am too tired to figure this out right now.  Bruising does not experience any sharp cutoff in the region I explored, and appears to follow the curve that deflection would have followed were it not subject to this cutoff. 

Edit:  Using the equation momentum = 50,000 / IMPACT_STRAIN_AT_YIELD give momenta of 156.7 (force=117.01) for iron (IMPACT_STRAIN_AT_YIELD=319), and 91.4 (force=68.2) for bronze (IMPACT_STRAIN_AT_YIELD=547).  This exactly agrees with the cutoffs I observe, so I think it is clear this is the same cutoff I observed before, and that 50,000 / IMPACT_STRAIN_AT_YIELD describes it accurately.  I will eventually have to test if it deviations from this equation for large IMPACT_STRAIN_AT_YIELD (corresponding to small forces/momenta).  Based on this equation, steel (IMPACT_STRAIN_AT_YIELD=940) should have this cutoff at mometum=51.2, or force=39.7 for copper bolts.  Given that this is low, it explains why I see bruising for steel, and not for any other armor.  I will have to check if the deflection drop actually does occur at 40 for copper vs. steel. 

I am still not sure exactly what's going of for the relationship between deflection force/momentum and IMPACT_FRACTURE.  As you can see, it's very linear, but both the offset and slope are not round numbers.  Converting from force to momentum (muliply by copper bolt mass=1.3395) does not make the numbers and nicer.  Using the old (nonuniform) "uniarmor", I found that IMPACT_YIELD plays at least some role in the offset, but not the slope, and IMPACT_STRAIN_AT_YIELD affects neither.  Maybe it would be easier to study this behavior looking directly at the stains using DFHack and modded skin?

I will try some more analysis of my results this evening (maybe the relationship tracks on the upper or lower edge of the transition or something) and will also post the spreadsheet of my full results to DFFD.  I also need to post a large backlog of results from the past few days, although some of those may be confusing/uninteresting as they use the non-uniform armor.

[Pirate Bob]

- For a given velocity, bolts impact with a range of possible strains which appear to follow a pattern. The strains are just the bullet hole in the wall, we need to learn about the bullet (the strain is an effect, not a cause).

What would happen if you made the target a more uniform creature than a dwarf, like maybe a sponge?  Something with just one body part that could be hit.  Would this remove/reduce the variation?

I have to try this myself sometime.  I assume you are using the same lua script used to read out bolt velocities to get this information?  Unfortunately this seems to crash my linux version of DF, so while I can see the report, I can only look at one shot per instance of DF before the game crashes.

I am also wondering if there is a relationship between the strain for fracture you found and the 50,000/IMPACT_STRAIN_AT_YIELD cutoff for deflections?

[Urist Da Vinci]

I'll start off by quoting myself:

...
See the plot below of strains in the steel skin at various velocities. The horizontal axis is the sample size, scaled to a sample size of 100. We see that bruising occurred 25% of the time at velocity 30, and denting the other 75% of the time, regardless of the body part hit. We see that fractures (only about a 5% fracture) occurred 23% of the time at velocity 45, and that is also regardless of the body part struck.

Spoiler (click to show/hide)

I'm not sure what to make of the stepped strains yet. It isn't related to contact or body part size, because both small and large parts were at all steps.

...

See in the image how the strain goes up with velocity in steps, and there are about 6-11 steps? Turns out each of those steps is linear with velocity, by itself. However the slope of the linear plot is different for each. The leftmost point (trailing edge) is represented by:
Strain = 76.1 * Velocity - 2350
and the rightmost point (leading edge) is represented by:
Strain = 94.1 * Velocity - 2350
with both results capped between 0 and "strain at fracture". The strain grows about 1.25x faster on the leading edge than on the trailing edge, so we expect to see sharp transitions at low velocity and gradual transitions at high velocity.

Now if we look at the distribution of "percent fractured" (i.e. %penetrated or crack length, 100% implies the tissue has been completely penetrated or destroyed) at a constant velocity, for iron bolts vs steel skin, with varying steel IMPACT_STRAIN_AT_YIELD:

Spoiler (click to show/hide)

We see that how badly your skin (armor) has been fractured doesn't depend on IMPACT_STRAIN_AT_YIELD. It also appears that the velocity-dependant stepping also applies to fracturing, so we expect to see sharp bruising-tearing transitions at low velocities and gradual bruising-tearing transitions at high velocities.

I then tried to find the exact values of IMPACT_STRAIN_AT_YIELD below which bruising stops for iron bolts vs steel skin:

Code: [Select]

Velocity  Nobruising  Bruising
35        1282         1283
40        1111         1112
48        925          926
55        806          807
60        735          736

This data can be represented by:
Strain = 48539 * (Velocity ^ -1.023)
OR perhaps by:
Strain = 1/(0.00002294 * Velocity - 0.00002035)


Access to the wound data has allowed me to figure out the "translation" for the different wound descriptions on tissue:

Spoiler (click to show/hide)

Shattering - the tissue has been 100% penetrated across 100% of the surface area of the body part, and the tissue material has IMPACT_STRAIN_AT_YIELD less than 25000.
Tearing apart - the tissue has been 100% penetrated across 100% of the surface area of the body part, and the tissue material has IMPACT_STRAIN_AT_YIELD greater than or equal to 25000.
Tearing - the tissue has some penetration, but it isn't complete or across the entire surface, and the tissue material has IMPACT_STRAIN_AT_YIELD greater than or equal to 25000.
Chipping - the tissue has been 100% penetrated across less than 25% of the surface area of the body part, and the tissue material has IMPACT_STRAIN_AT_YIELD less than 25000.
Fracturing - the tissue has some penetration, but it isn't complete or across the entire surface, and the tissue material has IMPACT_STRAIN_AT_YIELD less than 25000.
Denting - the tissue has not been penetrated, but it has more than 0 wound strain.
Bruising - the tissue has not been penetrated, and has no wound strain, but it is vascular (has blood vessels).

I haven't read your latest two posts in detail. The strain or fracture stepping happens even if you only look at one body part, and regardless of body part size. I'll probably reply later today.

[Pirate Bob]

I then tried to find the exact values of IMPACT_STRAIN_AT_YIELD below which bruising stops for iron bolts vs steel skin:

Code: [Select]

Velocity  Nobruising  Bruising
35        1282         1283
40        1111         1112
48        925          926
55        806          807
60        735          736

This data can be represented by:
Strain = 48539 * (Velocity ^ -1.023)
OR perhaps by:
Strain = 1/(0.00002294 * Velocity - 0.00002035)

This data appears to roughly fit momentum = 50,000/STRAIN_AT_YIELD.  This strongly suggests that armor "bruising" is the same transition I find for the sharp cutoff in deflection at momentum = 50,000/STRAIN_AT_YIELD. 

Edit:  Is it correct that this transition goes from zero to 100% bruising, with no intermediate regime?

Also, I haven't had time to test yet, but at least with the old non-uniform armor 50,000/strain wasn't exact (especially at low velocity/high strain) so I wouldn't worry that there's a slight difference. 

[Urist Da Vinci]

...
This data appears to roughly fit momentum = 50,000/STRAIN_AT_YIELD.  This strongly suggests that armor "bruising" is the same transition I find for the sharp cutoff in deflection at momentum = 50,000/STRAIN_AT_YIELD. 

Edit:  Is it correct that this transition goes from zero to 100% bruising, with no intermediate regime?
...

1. Note that it isn't the armor being bruised (in my cases the steel skin was being dented or fractured), but the fat/muscle beneath the armor.

2. Yes, either all bolt shots cause bruising beneath the armor (which if it hits the skull/brain leads to death) or all bolt shots don't cause any bruising.

...
What would happen if you made the target a more uniform creature than a dwarf, like maybe a sponge?  Something with just one body part that could be hit.  Would this remove/reduce the variation?

I have to try this myself sometime.  I assume you are using the same lua script used to read out bolt velocities to get this information?  Unfortunately this seems to crash my linux version of DF, so while I can see the report, I can only look at one shot per instance of DF before the game crashes.
...

3. Tested a giant sponge. It only has one body part and one material, yet I get stepped outputs for dent strains or penetration percentages.

4. Which version of DFHack are you using? I have dfhack-0.34.11-r2 and I got crashes as well when using dfhack-0.34.11-r1.

Spoiler: wounds lua script (long) (click to show/hide)

unk_10 is strain
unk_70 is % penetrated

Code: (wounds.lua) [Select]

unit=dfhack.gui.getSelectedUnit()
if unit==nil then
print ("No unit under cursor!  Aborting with extreme prejudice.")
return
end

for k,v in pairs(unit.body.wounds) do
print("BDWD")
print(k)
print(v)
for kk,vv in pairs(v) do
print("UNIT_WOUND")
print(kk)
print(vv)
if kk=="parts" then
for kkk,vvv in pairs(vv) do
print("PARTS")
print(kkk)
print(vvv)
for kkkk,vvvv in pairs(vvv) do
print(kkkk)
print(vvvv)
if kkkk=="unk_14" then
print("VECTOR")
for kkkkk,vvvvv in pairs(vvvv) do
print(kkkkk)
print(vvvvv)
end
end
if kkkk=="unk_24" then
print("VECTOR")
for kkkkk,vvvvv in pairs(vvvv) do
print(kkkkk)
print(vvvvv)
end
end
if kkkk=="unk_34" then
print("VECTOR")
for kkkkk,vvvvv in pairs(vvvv) do
print(kkkkk)
print(vvvvv)
end
end
end
end
end
end
end


...

Spoiler (click to show/hide)

This is for bronze.  Notice that for IMPACT_FRACTURE x 3, the deflection starts to go down smoothly, and then drops sharply to zero, while bruising continues to go down smoothly.  The cutoff in deflection may be the cutoff proportional to 1/IMPACT_STRAIN_AT_YIELD, but I am too tired to figure this out right now.  Bruising does not experience any sharp cutoff in the region I explored, and appears to follow the curve that deflection would have followed were it not subject to this cutoff. 
...

5. The pink and blue lines represent when the bronze armor has completely fractured, and the tissue underneath gets to be cut up by the bolt. Because "percent penetrated (a.k.a percent fractured)" comes in a variety of magnitudes for a given velocity (at several discrete steps), you get a smooth transition to tearing damage.
The yellow line represents the sharp transition to bruising damage. Remember that armor deflection is the absence of bruising, tearing, or chipping in the fat/muscle/organs/bones/etc. The armor (like a metal skin) is still being dented or fractured (it just doesn't output that combat text).

So default bronze armor fractures before bruising can happen, whereas default steel armor has bruising happen before the armor can fracture.

For other people who are following the thread: the way it is currently, if armor damage was tracked and armor could break, a single wood arrow from an elf could destroy a piece of adamantine armor by creating a gaping hole or shattering the entire armor. A wood arrow currently (in vanilla DF) can fracture adamantine armor and tear the flesh underneath, but the armor always acts as if undamaged:

The flying {mangrove arrow} strikes Dwarf 2 in the lower body, tearing the muscle and bruising the left kidney through the small adamantine breastplate!

[GlyphGryph]

Zivillin, could you perhaps add links to supplemental posts in this thread from the OP? There's quite a few posts here already, so it's hard to review the additional research to see if my question has already been answered.

You said you tested against dwarves with no skills, and I believe Pirate Bob did the same, and thus Plate Armor was ineffective.

I would like to know if maxing out armor skill and dodging and giving them open space changes that - it could be that armor like plate provides a multiplicative bonus based on armor skill in situations like this (increasing deflections) that is effectively canceled out by the total lack of skill. 200 x 0 is still 0, after all.

If this is already done, all the better, but I suspect it may be like shields, where the skill involved is one of the more important (if not dominant) factor, since it's simply "does it deflect, or does it not deflect"

[Pirate Bob]

Zivillin, could you perhaps add links to supplemental posts in this thread from the OP? There's quite a few posts here already, so it's hard to review the additional research to see if my question has already been answered.

You said you tested against dwarves with no skills, and I believe Pirate Bob did the same, and thus Plate Armor was ineffective.

I would like to know if maxing out armor skill and dodging and giving them open space changes that - it could be that armor like plate provides a multiplicative bonus based on armor skill in situations like this (increasing deflections) that is effectively canceled out by the total lack of skill. 200 x 0 is still 0, after all.

If this is already done, all the better, but I suspect it may be like shields, where the skill involved is one of the more important (if not dominant) factor, since it's simply "does it deflect, or does it not deflect"

I performed all armor tests using dwarves with grand master armor user (the max available in the arena).  I don't see how giving them the ability to dodge and dodge skill would change the results, but it's an interesting idea so I will try it when I get a chance.   

I think there is just too much stuff here to even put links to all of it in the OP.  Our plan is to wikify it eventually, but for now I am working on a google doc with much of the information (although it is not up to date).  That reminds me - I need to put more information on the conditions used for testing in this document.

1. Note that it isn't the armor being bruised (in my cases the steel skin was being dented or fractured), but the fat/muscle beneath the armor.

Oh, well, then it's the exact same thing I was looking at, so that makes sense that you would get the same result.  That's great then.  I've got quite a few points on this transition, and, like I said, for small STRAIN_AT_YIELD, it goes like momentum=50000/STRAIN_AT_YIELD, but at very large STRAIN_AT_YIELD (say maybe >1000) the needed momentum deviates slightly higher than this.  I have to confirm this deviation still occurs with truly uniform armor.

Edit:  I tried the latest version of DFHack, and no crashes running your scripts any more, so as you suspected it is a version issue, not a linux issue.  Now I have to think about combining it with my scripting...

[Urist Da Vinci]

This appears (I don't know why adamantine is different, perhaps my tests there need to be peer reviewed) to work at predicting the velocity required to completely fracture armor (or metal skin) of standard (low) thickness:



This function doesn't show contact area or thickness yet. I think Pirate Bob found earlier that the round-to-100 rule applies to the contact area and thickness scaling for this function. I used a fudge factor of 2 to make the results work in the case where contact area was larger.

"Completely fracture" means the velocity at which you see the smooth sigmoidal transition in deflection and appearance of tearing damage.

[pisskop]

ptw

[Pirate Bob]

This appears (I don't know why adamantine is different, perhaps my tests there need to be peer reviewed) to work at predicting the velocity required to completely fracture armor (or metal skin) of standard (low) thickness:



This function doesn't show contact area or thickness yet. I think Pirate Bob found earlier that the round-to-100 rule applies to the contact area and thickness scaling for this function. I used a fudge factor of 2 to make the results work in the case where contact area was larger.

"Completely fracture" means the velocity at which you see the smooth sigmoidal transition in deflection and appearance of tearing damage.

While this equation is definitely a start, it is does not fully describe the results from my last post for copper bolts vs. iron and bronze at varying IMPACT_FRACTURE.  I found that for iron armor, momentum=IMPACT_FRACTURE/22436-12.453 and for bronze, momentum=IMPACT_FRACTURE/22294-15.426 (multiply the fits for forces by copper bolt mass=1.3395).  While the coefficients of IMPACT_FRACTURE are close to 22500, the difference is significant.  Also, the offsets are really not close to those predicted by IMPACT_YIELD/45000. 

1) How do you define the location of the transition?  I define it as the force/momentum where non-bruise wounds occur 50% of the time.  Maybe you (and/or Toady) are actually defining it based on some other threshold?

2) Shouldn't your equation strain at fracture = IMPACT_STRAIN_AT_YIELD * IMPACT_FRACTURE / IMPACT_YIELD describe this same transition?  (Edit - it appears that is actually something different, and the equivalent thing would be the momentum/strain needed for 100% fracture?  Can you confirm this?)  If so, then we should expect the momentum at fracture also be a function of IMPACT_STRAIN_AT_YIELD * IMPACT_FRACTURE / IMPACT_YIELD.  It's possible that some factors (like 1/IMPACT_FRACTURE, for example) might cancel out, but it's also quite possible that the final equation becomes very complicated (which I think is likely given that simple linear fitting doesn't seem to quite work).  Also, how sure are you of the accuracy of strain at fracture = IMPACT_STRAIN_AT_YIELD * IMPACT_FRACTURE / IMPACT_YIELD?  Should I try to confirm this over a wide range of impact parameters?

3) Have you done anything more towards figuring out what determines the strain on the material?  Assuming (2) is true, this would be extremely informative for figuring out armor protection.  I could do this myself now that you got me to use the right version of DFHack, but it seems likely you've already been working on this.  If I was going to do it, I would imagine setting SHOOT_FORCE such that the strain was about half of its maximum value, and then seeing how the strain changes when different parameters are varied.  This has a huge advantage over deflection testing, as you get an exact numerical output of how the strain changed from a single hit(Edit - the result is not really "exact", as there is variation in the strain you get at a given bolt momentum - not clear if there is actually a great advantage over looking at deflection).  In other words, if (2) is true and you don't plan to do this, let me know and I will try.

I really, really need to put up my full data for all materials at contact area 2 and 10.  I'm sorry I didn't do this earlier, as it would have helped you a lot when working on the relationship between momentum for injury and impact parameters.  There's just a lot of it and I'm not 100% sure the best way to represent it.  I'll try for a spreadsheet of the raw data on DFFD if nothing else.  If you could tell me what % serious wounds you are looking at for your equation that would be helpful, as then I could make a table of that.  I am also running more simulations varying IMPACT_FRACTURE with IMPACT_YIELD set to zero right now, which hopefully should shed some light on the relationship between momentum required to cause injury and the impact parameters.

Edit:
I looked at the data for IMPACT_YIELD=0, and this data perfectly fits to momentum=IMPACT_FRACTURE/22500 for iron, bronze, and adamantine over the whole range from 1 to 6 times the vanilla IMPACT_FRACTURE.  The fit is perfect in all cases to within the error of the measurements.  This means that the offset and the corrections to the slope must be some function of IMPACT_YIELD, possibly multiplied with other factors (like IMPACT_STRAIN_AT_YIELD). 

Spoiler (click to show/hide)

I have posted the complete results for this and varying IMPACT_FRACTURE at vanilla values of IMPACT_YIELD for iron an bronze on DFFD.  And it is late, so I need to go to bed.  Sorry, I will really try to post my older data soon  .

[Urist Da Vinci]

...

1. I wasn't that consistent on the % deflection. I trust the iron/copper on steel results a bit more than the copper on iron/bronze ones because the material properties for copper, iron, and bronze all overlap somewhat:

Spoiler (click to show/hide)

Perhaps zinc bolts would produce better bronze armor results?

2. Nope, "strain at fracture = IMPACT_STRAIN_AT_YIELD * IMPACT_FRACTURE / IMPACT_YIELD" is a very hard result and you can see it for yourself using the wounds script. The strain for cut leg bones (i.e. use an adamantine sword to stab the leg) is 113, which is roughly equal to 100 * 130000 / 115000. The strain at fracture for shattered bones is 100 since the impact yield and fracture are identical there.

Also see what I posted earlier:

...
Now if we look at the distribution of "percent fractured" (i.e. %penetrated or crack length, 100% implies the tissue has been completely penetrated or destroyed) at a constant velocity, for iron bolts vs steel skin, with varying steel IMPACT_STRAIN_AT_YIELD:

Spoiler (click to show/hide)

We see that how badly your skin (armor) has been fractured doesn't depend on IMPACT_STRAIN_AT_YIELD. It also appears that the velocity-dependant stepping also applies to fracturing, so we expect to see sharp bruising-tearing transitions at low velocities and gradual bruising-tearing transitions at high velocities.
...

3. Perhaps instead of plotting deflection results (which can be misleading as there are at least two reasons why a bolt is no longer deflecting), plot the %bruising and %tearing. It'll start out near zero and then either bruising or tearing will suddenly increase as you increase velocity. We don't have to see the percentages for other types of injuries, as we are just looking for certain cutoffs.

If the momentum isn't high enough to initiate fracture, the resulting strain is randomly chosen from the stepped results, but overall appears to scale linearly with momentum (each step scales at a different rate, so the spread gets wider at higher velocities!). I don't think it's worth the effort right now to research in that direction.