You are on page 1of 11

http://ftp.lstc.

com/anonymous/outgoing/jday/concrete/mat72r3

FYI, this report described a study pertaining to validating the capability


afforded by
the standard release of the Karagozian & Case (K&C) cementitious (or
concrete)
material model. In doing this, responses computed with the KCC-model for
structural components comprised of either concrete or reinforced concrete
were
compared with results from tests. Comparisons of the responses computed for
the
structural components were shown for both quasi-static and blast loadings.
Computed responses were also compared to data from standard material tests
to
demonstrate the basic capability afforded by the KCC-model. The analytic
results
shown were computed with the LS-DYNA code. The KCC-model is embedded in
this code
(i.e., Release III, Model MAT072R3 in LS-DYNA):

Crawford, J.E., Wu, Y., Choi, H.-J., Magallanes, J.M., and Lan, S., "Use and

Validation of the Release III K&C Concrete Material Model in LS-DYNA,


TR-11-36.5,
Karagozian and Case Structural Engineers, Burbank, California, July, 2012.

http://www.kcse.com/assets/Uploads/TR-11-36-6-KCCM.pdf

Best,
Jim Kennedy
2/17/14
_____________________________________________________________________

We concluded that R7 is indeed the correct code implementation of MAT072_REL3. The R6 version is consistent with R7 IF all
the inputs fields are input directly (i.e., the “direct input†option); however, there is a bug related to the
“automatic†parameter generation option. In this R6 code and the “automatic†option, where most of the input card is
typically set to zero, a line in the code that resets the b1 parameter to a default value was commented out; consequently,
the R6 code was using a b1 value of 0 for all of “automatic†parameter option. This is the reason why I was so confused
at the differences I was seeing between R6 and R7 for the problem I was handed to look at; here at K&C, we rarely use the
“automatic†option (because I have a separate code that computes values for different types of concretes and
concrete-like materials and input all the parameters directly).

Let me know if this clears this issue up.

Lastly, I still think there are some peculiarities between the Windows and Linux versions for R7, but I haven’t had time to
put this in a format for you. I’ll see if I can reproduce some of those peculiarities and if I see them again I will send
them over.

1 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

Joe
Magallanes@kcse.com
11/22/13
__________________________________________________
bugzilla 8972 describes bug affecting
R60 and R61 (prior to 9/26/13).
____________________________________________________

RE: erosion
There is no erosion criterion in the material model. Can use *mat_add_erosion.
__________________________________________________________________________________
Bug #8512, making *mat_072r3 unstable when *damping_part_stiffness is applied, is fixed
in v81304dev, R70.81305 on 5/2/13 (LL).
__________________________________________________________________________

The sample curve tabulated in the User's Manual has strain rate units of 1/millisecond whereas your curve's abscissa should
be in units of 1/second. Using 971 Dev, the messag (or mes0000) file will include a full set of material input parameters,
generated internally based on the unconfined compressive strength provided by -A0.

If NOUT=2, the "plastic strain" in the output actual represents the scaled damage measure (per Table 72.1 in the latest
User's Manual). This damage is 1.0 when lambda = (lambda)m, signaling arrival at the maxium failure surface. As lambda
increases from that point, the failure surface transitions between the maximum failure surface and the residual failure
surface. Damage=2.0 signals arrival at the residual failure surface.

The latest User's Manuals are always available at www.lstc.com/download/manuals .

____________________________________________________
we did try to let NPTS be greater than 13 to help regularization, but
it is far away from mature, so we don't want to release it and let public
users to try it now (that's why we didn't say anything about it in the Users
Manual).
So, my answer to your question is: NPTS<=13. Although the Users Manual says
it must be 13, the code actually allows NPTS<13. But when user uses
automatic generation, it will always be 13.
Youcai
6/6/12
__________________________________________________
RE: output of parameters in convenient form to messag when automatic
parameter generation is used.
This was eariler disabled because of the unclear warning
$ >>> Users need to change/check: MatID & RO & Rsize & LocWidth for units <<<"

(Leslie incorp. below in v73017dev)

Leslie:

Please use this code [attached to email].

2 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

Note that parameter “ro†is added to the argument of subroutines “prmt72†and “prmt72r3†. Therefore, you need
to add “ro†to the call statement “call prmt72(…)†in subroutine “printm†.

With these changes, the output “messag†file is attached for the corresponding input “mat72r3.g-mm-ms.k†. By doing
this, users can copy the keyword part of “*MAT_DAMAGE_CONCRETE_REL3†and “*EOS_TABULATED_COMPACTION†and make their
own input easily.

Now, RO, RSIZE, and UCF are the ones defined exactly by users in the automatic generation version, which is required!

However, the MATID and EOS ID are purposely set to 72. The reason is, users have to change the *PART card when direct input
is used (EOS ID must be input). Users are forced to change the MATID and EOS ID on both *PART and *MAT cards so that they are
consistent and this CANNOT be done without any manual operation. Nevertheless, MATID and EOSID are just some arbitrary unique
numbers that have nothing to do with the material properties.

Youcai
KCSE
3/28/12
___________________________________________________________________
In my previous work, I did quite a bit of work with seismic loads, so I do appreciate the need and difficulty of the seismic
problem. We've also discussed the concrete model with Bureau of Reclamation for some time now (in fact, I think we gave them
a copy of the source code for an older version some time ago, but I am not sure). Most recently, Lawrence Livermore National
Laboratories had an interest in seismic as well, but their interest had a slightly different twist.

To answer your last questions:


(1) "Yes, there is compression strength after tensile failure with confinement, even higher than the unconfined peak
compression. Is this because the damage parameter does not go to 2 even though the tension stress is zero? The confined
monotonic compression peak is almost 6000 psi (same confining pressure)."

- The reason for this is that fractured concrete has a residual friction when placed under confinement, much like a sand or
other cohesionless geomaterial will. This is encapsulated in the model with one of the three failure surfaces. When you are
damaged closer to "2", the model interpolates toward the third "residual" strength failure surface; the higher the
confinement, the higher the residual shear strength, which is consistent with data.

(2) "One thing that seems odd is that the slope to failure in the softening range of compression is abrupt, almost vertical,
but that of the tension is flatter. This holds true for both unconfined and confined. It seems it would be the other way
around, unless I'm missing something. Any comments on that?"

-I'm not sure I see what you mean. If you mean that the there are differences in the post-failure response in compression
versus tension, this is just a function of several parameters that a user can vary if your data tells you otherwise. For
blast, these "default" values work remarkably well, but you can change them with some experience using the model (see next
question).

(3) Is there any way to tweak the stress strain relationship to match test samples by plugging in some of the parameters
rather than using the defaults?

-Good question. Short answer is yes. The long answer is that I have developed a scheme consisting of a little art plus

3 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

science to do this which is difficult to explain in an email. I am attaching a recent paper I put together that touches on
this subject. See the section "Parameter Estimation Methodology".
(see http://ftp.lstc.com/anonymous/outgoing/jday/Magallanes_et_al_2010b.pdf )

There are several avenues to take to attempt using the LS-DYNA concrete models for seismic applications, but I would
definitely recommend V&V (verification and validation), since this is really an area that has not been explored with fully 3D
continuum simulations. I can send you a technical report we put together that has a few example V&V seismic problems that we
did several years ago if your interested. Bottom line is that the blast/impact concrete models including MAT072R3 is a very
good option for this but there may be inevitable improvements needed to work your problem correctly. I'd be happy to discuss
this, maybe even put some other people together in the Government who would be interested in this, since it does seem like an
important application to cover. I'm traveling for the rest of the week (starting tomorrow), but am in my office if you would
like to set up a time to discuss (818-240-1919).

Joe Magallanes (in email to ggauthie@water.ca.gov)


11/29/11

Based on test data I have seen, the following occurs during stress reversals (again, limiting to unconfined cases, i.e.,
uniaxial stress):
(1) Compression into the post-failure softening regime, then reversal in tension. After being heavily damaged in compression,
there is very little tensile capacity. Test samples are faulted (at some inclined angle) if not heavily micro-cracked.
(2) Tension into the post-failure softening regime, then compression reversal. The tensile failure will generate a crack
plane perpendicular the the applied loading direction. Upon reversal, and after closure of that crack, there would be some
resistance to compression, albeit at some reduction from its pristine capacity.

MAT072R3 handles these two cases as follows:


(1) Softening occurs when damage is between 1.0 to 2.0 (1.0 designates the maximum strength, while 2.0 is the maximum damage,
as indicated by Jim, which softens the material to the residual failure surface). Since R3 uses one damage internal variable,
tensile reversal stresses will be less than the pristine tensile strength. If the element is damaged up to 2.0 in
compression, then the tensile strength upon reversal with be effectively zero. Although not perfect, this is consistent with
test data.
(2) When completely softened in tension (2.0), the compression stress will similarly be zero, effectively. I think Gary's
point that this is unrealistic is well-taken, but this will only be true in the unconfined case (uniaxial stress). If you
have some confinement (for example due to steel reinforcement or from the inertia of surrounding concrete), you will see that
MAT072R3 will indeed have some compression resistance that is less than the maximum strength capacity. So again, not perfect,
but consistent with data.

This all being said, these issues are not unique to MAT072R3. I know at least two of the other concrete models in LS-DYNA
would yield similar results. Most of those models were originally developed for blast/impact applications, where stress-
reversals were not the primary interest. I don't know of much data that is available from which to calibrate reversal effects
(I'd be happy to look at it if Gary and all have some). I also don't know of very many 3D continuum plasticity models that
have been specifically developed/validated for seismic applications (would this be Gary's primary interest?), although our
new version may better handle that (MAT072R4). I hoping to get that over to Jim and LSTC soon.

Joe Magallanes
11/29/11

4 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

_______________________________________________________________________________
For better understanding of *mat_072r3, See these references ...

- Malvar (1996), http://ftp.lstc.com/anonymous/outgoing/jday/concrete/concrete.CP27_WES96K.pdf.

- Schwer, L.E., and Malvar, L.J., "Simplified Concrete Modeling with *MAT_CONCRETE_DAMAGE_REL3," 4th German LS-DYNA Forum,
Bamberg, Germany, October, 2005.
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat_072r3.pdf.

- "Recent Improvements to Release III of the K&C Concrete Model", Magallanes et al, http://www.dynalook.com/international-
conf-2010/Simulation-1-4.pdf .

- "K&C Concrete Material Model Release III—Automated Generation of Material Model Input, Malvar et al, 2000."
(Contact Joe Magallanes at magallanes@kcse.com to request this one.)

For further information on concrete modeling , see the files in http://ftp.lstc.com/anonymous/outgoing/jday/concrete,


starting with the file "concrete_models_faq"

________________________________________________________________
From correspondence with Len, 11/2010:

2. Perhaps that line in the code:

ywhrt(i) = max(ywhr(i),ywhrt(i))

covers the case where the user inputs the DIF values in the wrong order,
e.g. the DIFs for the negative entries (tension) are smaller than the DIFs
for the positive entries (compression).

That is your Define_Curve 99 case

3. The response is linear elastic if the initial yield has not been reached
AND the mean stress is below the transition to the void collapse pressure,
i.e. the linear elastic bulk modulus is operative.

4. Youcai Wu of K&C just mentioned Releases 4 in response to a question I


asked at the meeting.

Len

-----Original Message-----
From: Jim Day [mailto:jday@lstc.com]
Sent: Saturday, November 06, 2010 9:28 AM
To: Len@Schwer.net

5 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

Subject: Re: Mat_072R3 with rate effects; Release 4 of K&C model

Len,

2. Good point about my test curves 8 and 9. In the attached model, I


reconstituted curves 8 and 9 as curves 88 and 99 wherein the enhancement
factor is 1.0 at zero strain rate. Using curve 99, I still find that for
tension,

peak tensile stress/nominal (unenhanced) tensile strength = 7.6/3.8 = 2,


with the conclusion being that

tensile strength enhancement


= max [tensile rate enhancement from curve,compressive rate enhancement from
curve] = max (1.5, 2.0] = 2.0

See http://ftp.lstc.com/anonymous/outgoing/jday/concrete/m72r3.rate_effects.k

The source code for 72r3 appears to confirm it ...

...

3. Is the response of the material EVER linear, even in the low tensile
strain and in the low compressive strain regimes?

4. Are there any publications that speak of the enhancements of Release 4?

Thanks,
Jim

Leonard E Schwer wrote:


> > Hi Jimmy --
> >
> > I want to understand your reply.
> >
> > 1. Your conclusion is that negative strain rate values in LCRATE
> > correspond to tensile states, e.g. when the mean stress is negative. I
agree with this.
> >
> > 2. I did not run your nice example, but I do not understand the caveat
> > you
> > mention:
> >
> > "There appears to be one caveat, i.e., tensile strength enhancement =
> > max [tensile rate enhancement from curve, compressive rate enhancement
> > from curve]"
> >

6 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

> > The only non-magnitude difference between LCRATE 8 & 9 is in the
> > former the zero rate value is the same as the compressive value, and
> > in the latter the zero rate value is the same as the tensile DIF.
> >
> > So why doesn't LCRATE 8 & 9 provide the same caveat?
> >
> > I also suggest you do not use anything other than UNITY for the
> > zero-rate value in any LCRATE - this is not physical and I am unsure
> > if the algorithm implementation depends on the unity at zero rate
> > assumption; programmers do 'odd' things :)
> >
> >
> > 3.If a user wants to adjust the elastic modulus, they will need to use
> > the full input version of MAT072R3.
> >
> >
> > Finally, last week I attended a session at the recent Shock &
> > Vibration symposium where one of the K&C guys admitted they now have
> > Release 4, but have not provided this releases to LSTC.

_____________________________________________________________________
Yocai from Karagozian and Case reported seeing bad behavior of
mat_072r3 when using solid ELFORM 2 and also when doing implicit analysis.
He will contact John (6/27/08).

I found a mat_072r3 example with solid elform 2 in our QA suite. Even


in this QA problem involving penetration of mat_072r3 material, the
elform 2 response is dubious as compared to the elform 1 response. On
the basis of what I've seen in the test cases, I would say the prudent
recommendation is not to use ELFORM 2 for mat_072r3.

Alyotech also reported bad behavior of ELFORM 2 in reinforced concrete beam


test case.
_____________________________________________________________________

The information echoed to the 'messag' file includes the following:

$ >>> Users need to change/check: MatID & RO & Rsize & LocWidth for units

This warning was to indicate the values were hardwired, as these values
were not readily available in the subroutine that wrote the 'messag' file
information. Perhaps I should have been more explicit when writing this
warning, or left the entries blank.

--len

_____________________________________________________________________

7 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

From emails 9/22/06 thru 9/25/06 (home machine)

Thanks Javier.
[Your comments are worth much more than 2-cents! :)]

#1 - I mis-read Jim Day's plot, the scaled damage parameter did NOT exceed
TWO. So all is OK with this part of Mat072R3.

#2 - Jim - are you aware of LSTC receiving updates from K&C?

Thanks again Javier, --len

-----Original Message-----
From: malvarl [mailto:malvarl@adelphia.net]
Sent: Sunday, September 24, 2006 9:34 PM
To: Len@Schwer.net; 'Jim Day'
Cc: torodd@lstc.com; 'Jensen'; 'Mediavilla Varas, J. (Jesus)'
Subject: RE: Nonlocal and mat72_r3

Hi Len

My 2 cents:

#1 - Since the scaled damage parameter = 2L/(L+Lm) where L = lambda, and Lm


= lambda at maximum failure surface, epx1 would tend to 2, but should not
really reach it, nor exceed it (it is 1 if L = Lm).

#2 - In quasi-static runs, the failure using Mat72R3 will not diffuse, but
it should expand the same amount of fracture energy. Unfortunately, there
is a bug in the current LS-DYNA version which makes this happen only if the
mesh has uniform elements - this has been corrected and will be passed on
to LS-DYNA.

Thanks
Javier

-----Original Message-----
From: Len Schwer [mailto:Len@Schwer.net]
Sent: Saturday, September 23, 2006 12:31 PM
To: 'Jim Day'
Cc: torodd@lstc.com; 'Jensen'; 'Mediavilla Varas, J. (Jesus)'; 'malvarl'
Subject: RE: Nonlocal and mat72_r3

Hi Jim --

I spent sometime looking at this problem and have two observations.

8 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

#1 - I think the problem as posed by Jesus of a fixed-end rod under step


tensile loading is not suitable for examining the effect of *Mat_NonLocal.

The fixed-end causes the tensile wave to double and this fails the element
closest to the fixed end. I doubt refining the mesh will change this
results, as this is what the physics also predicts.

I guess it was hoped that *Mat_NonLocal would "smear out" this end failure.
Apparently it cannot in this case, which in my assessment is the right
thing; but as always I could be wrong about this and everything else.

BTW - the scaled damage parameter, epx1 in this case, should range between
ZERO (no damage) to UNITY at the failure surface and then TWO when the
residual failure surface is reached. Why values greater than TWO are
reported in this problem is unknown; perhaps Javier can comment on this?

#2 - To see if *Mat_NonLocal is working with Mat072R3, I modified Jesus'


problem to be quasi-static with a prescribed displacement rather than a
stress end load. [There is still perhaps a bit too much dynamics in this
simulation, but I get paid by the hour so I do not run freebie jobs for too
long :)]

I also introduced a 'random' mesh scale along the length of the rod. The
function of *Mat_NonLocal is to prevent strain from accumulating in the
small (shortest) element. So if the mesh is uniform, the strain should not
accumulate in any one place and hence such a uniform mesh is not a good
test of *Mat_NonLocal.

I ran the above described (and attached) model twice:

Case 1 - omitting *Mat_NonLocal


Case 2 - with *Mat_NonLocal setting L=100 [nominal cube root of element
volume is about 36=(20x50*50)^(-1/3)]

When *Mat_NonLocal is omitted, Case 1, the failure (strain) localizes in


Element #49, which is the second shortest element length in the rod; the
relative element lengths can be observed in the d3hsp file by looking at
the list of smallest time steps.

For Case 2, with *Mat_NonLocal, the failure is diffused over several


elements; mostly in the vicinity of Element #49. It might be worthwhile to
run this implicit/static to see what (really) happens?

9 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

So I think *Mat_NonLocal is working with Mat072R3.

The troubling part to me is omitting *Mat_NonLocal SHOULD also diffuse the


failure as Javier has a built-in localization limiter in the parameter
LocWidth(=100 in this case). Perhaps Javier can comment on this?

--len

-----Original Message-----
From: Jim Day [mailto:jday@lstc.com]
Sent: Friday, September 22, 2006 12:04 PM
To: Len@Schwer.net
Cc: torodd@lstc.com; Jensen
Subject: Re: Nonlocal and mat72_r3

Len,

Just looking at the fine mesh problem, *mat_nonlocal definitely affects


epx1 ("effective plastic strain" which is supposedly the scaled damage
function) of some elements but such that epx1 is not monotonically
increasing (see attached jpeg). Make any sense to you? Also, should
epx1 be in the range of 0 <= epx1 <= 1.0?

Jim

Mediavilla Varas, J. (Jesus) wrote:

>> Dear Jim,


>>
>> I did the 1D simulations as Len suggested. Unfortunately I do have
>> mesh dependent results. See attached figures.
>>
>> After reflexion on the fixed end, strain localizes in one element.
>>
>>
>> I have tried as nonlocal variables, NL=1 and also NL=7, with the
>> damage scaled variable NOUT set to 2.
>> I'm sending you the indeck files corresponding to three different
>> meshes. Perhaps you can take a look.
>>
>> I have switched off the strain rate hardening, because this might (and

10 of 11 25-Aug-15 0:18
http://ftp.lstc.com/anonymous/outgoing/jday/concrete/mat72r3

>> it does) regularize the problem.


>>

11 of 11 25-Aug-15 0:18

You might also like