GeoSci.xyz SimPEG, 17 Jan 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SimPEG meeting Jan 16

Description

Seogi Kang talks about SP and Dam safety.

A

B

Oh, it's not sharing any more.

C

D

All right, look at you not your screen share. Oh.

C

It changes when you plug in I've had that issue sure.

A

So now, what is it bricks.

B

Yeah, our can I start absolutely.

A

I'm not prepared to teach, you have a sub thought so does this is the presentation that I gave in a GU with the sunken and down so the his problem is I've got that monitoring. So here's our setup. We got this earth dam. The left side is the reservoir, so we have some cod with glue and then the other side. It's like they're pop some people leave, and so it's very important to know where there's a wicked or not and oldie like hydraulic head should go down below the land.

A

So that's the problem and then the structure that they have a sand. So it's like a it's all that like a solid wrong, but did you yeah.

E

A

B

A

Special, so here's a sane, so that's basically at the loose sand and then there's a kite core. It's a click or like it's very important.

C

A

How do we had real, like dressed, like it, kinda, go down pretty quickly in this song, so it's like I had a really resisted layer or the resistive target in the DC problem.

A

So yeah like the motivation to use SP something like that, because you can still use the C problem because it sees it sees the conductivity contrast and water saturation. It will be different between like kind of is fully saturated zone and then dries well. So, if that, like changes due to the leakage using the DC, you can still see something, but still that's not like a direct detection.

A

For instance, if you have a linkage here, only changes you can see is like that: its water height difference because that changes the water saturation but like sometimes fully look like a water, is saturated fully but like it doesn't let you know how fast it moves. So how many little much wickers you have it's kind of hard to see from the DC problem. If you want to see with the DC problem, is a huge leakage. So what's the alternative?

A

Sp could be an alternative, because it's a more it's more direct method, because what way it does like, if there's a flow, that flow is like that's the source of the spcl. So it's a little bit different from DC. Now you don't have any control source, you got. You put the same like a electrode arrays, but like this water flow generates the electrical signal because it moves the ions and you can measure it on surface yep.

F

So, thank you very much like the school doesn't evoke you. What is rated no.

A

That's the mica, it's the water level, so that's like this is, will be a dry zone. This can be fully saturated.

A

uh It'll be like higher I guess, Oh.

D

Higher than the left yeah.

G

A

um Right I thing: cuz: it drops quickly so like if you don't having liquors. It goes a little bit down like this.

G

Equals zero on that right, exactly yeah.

A

Okay, so this is sort of henan fat conceptual model and that's actually the fluid flow. This is a velocity of the fluid and the his.

H

Rid of the top of your morality, Jaime, thank you. Have that yellow get rid of this off the other issues, so.

A

Like it's a little bit fate that I so yeah.

H

A

I've set up a boundary condition at here as a bit of judicial and then I put a move on until here, but all those boundaries. So there's a little bit of details because I had to define the boundary condition inside of your match. So I'm, actually not solving like this part, but not only solving inside of your mesh, is.

H

G

H

The texture finish this.

G

Is your note yet yeah.

E

It's a little bit annoying. It's.

G

Unnecessarily long, this lessens nothing in the amount of water. Just tell you get. There was water, faster gray,.

A

Right right so yeah here, I think what I did is actually I. I I've solved this full domain compute the pressure while I should solve for hydraulic head and then you can compute the pressure by subtracting the height. Then you find like a zero like a zero contour, so I'm assuming above this one, it's zero pressure, its meaning like there's nothing happening that like yeah. It's not quite right.

A

Ok, so idea was actually like a developing a simple of line tool because, like source of DSP is the fluid more like a toy flow and then we're getting the electrical signal. So you need this full I. Guess software's need some sort of food modeling that you can simply compute the hydraulic head and a hydric velocity, and that's your source of SP problem is DC and solve this SP.

A

That is a goal and then, at the end you may want to recover some hydraulic parameters of the earth that you can see for it. So that's what's the goal so by RC flow is basically DC equation, very assuming the state state and on the assumption that I made was k's not depend upon the saturation or how you look at, which is not a very good idea, but quite sometimes fine, I think this earth. That example is not a very good example to do that, but I did so.

A

I usually put the three shui on each side and then like for the top for the month and.

C

When you say that, because your mesh is actually like the staircase right.

B

C

That's on these faces, yeah.

A

C

Enjoyment and then these fries.

A

So you need a little bit of.

A

Kind of those types of like you need to think about that, and you need to like do this kind of things that a little bit of like utilities.

H

A

Does it um yeah just I, don't know, I was a big fan, cuz I yeah there's a better way: I either used like I, don't know like it unstructured nation' or something like that, which is better I.

B

Didn't have it so I.

A

Had to do that, oh yes, so there's a details so and yes, it's a little bit annoying but but like I, feel if you see the end product, that's actually basically same as how you solve a typical.

G

A

Like octree so like how it looks like in your.

G

Code, that's felt similar, so air cells inactive or is that trying to match, as that's all cells that you have any entire problem, is how you can organize it. So.

A

I'm, actually only solving inside of this that, like a blue line, I'm not solving above I, mean time it generate the mesh yet, but on you, I, don't like I. Just get excited yeah cut it off, but you need to be a little bit careful about the boundary condition when you're doing that, because you need to find all the corresponding like a faces and then define your longer conditions.

A

Yes, oh that's wit, bc, resolver, a missile Darcy problem and you can compute the fluid velocity or having a cat, which is will be the source of the SP problem and well then, once you get the fluid like that property, you need to solve the sea problem. So we got like this step, a statement or equation.

C

A

Know really well there's a little bit of difference. Now, your total current, the summation of your streaming current plus conduction currents, so conduction currents is what we usually think. The sea currents- and we know like this- should be continuous, so DJ should be zero if we.

C

Don't have any like.

A

That active source, so by taking this and then we can have this equation so DJ s is equal to minus. The deep GC is equal to minus d pjs. That gives a little bit of physics, because your conducts direction of your conduction current is always opposite your streaming currents. So if the conduction current flows like that, but the streaming currents flow is the other way.

A

Okay, if you want to know a little bit more of a there is like a full book like explaining. Why is that we're like that sort of simple you're.

C

In the the conduction current, when you say conduction current, that is like ah yes,.

A

That's following orders: no.

D

The conduction current is.

A

Actually, a loser, moving charges, ray meters.

C

Yes, so that would be due to the floor. Water right.

A

Yeah well like it's it's induced by this, but like it exists together coincidence, but the like is two different like basically, there are sort of different currents, though, because that like then that current should be zero right like a diff jas diff j is 0. There too, I guess, there's conduction cons and also streaming currents exist, um ones.

G

Like the flow of ions in the water, which is jas, and one would have to be some movement of a letter to charge em that right balance yeah exactly oh.

F

A

They're opposite this I think that's important to them. Then Landeros like this, like a typical equation that we know so like this. This jess is different from what we usually use as an active source, rusev delta function, but yeah anyway, we can use it as same notation. Then. Basically, it's the same DC equation. So now you need to know what is then what is creaming currents? So we define streaming currents as like this. So it's like a different constitutive relationship. Your L is the that's an important, that's the key parameter which call cross coupling coefficient.

A

So it's actually how much potential electrical potential is generated due to the hydraulic head difference so like it's Delta, Phi Phi by Delta, H. So and then it's proportional to Sigma and there are a couple different ways: you can define that as like I, don't like headwear hydraulic velocity, so something like there's different people use different terminologies, but at the end of the day, like that's basically same equation with that different terminologies.

A

G

The earth, material and conductivity students move just have differentiate yes, I laughing. That Sigma is all all those their home together players taking into account conductivity of the yeah.

A

Well, this is like a mop free, nonlinear property, and this is unknown. So, like people like given in lab scale, they were trying to know what it's like l and they were like doing some experiment, they're quite a few papers, but like talking about that yeah yeah, it yeah, it's like it's like a dump that I'll be it's all. This, like a effect happening, is that if you think about that, that's pretty nonlinear right, wait. Yeah.

G

Cuz, then also, if you think back.

A

About the flip problem, that is, that was pretty nonlinear and then nonlinear airy, no Natalie actually propagate back to this L. That's.

B

A

Yeah then, why we're thinking of? But the problem is very simply. Okay, let's say you know the AL.

A

B

A

Then yeah, so one interesting thing you can think about is like. If you kind of like a decomposed this chair, then you can break apart as a two charms j/s was like QV x. You ok, sir.

C

A

Is like a it's also, not its excessive electrical charge per unit volume, so it's pretty similar to l but like units, it's a little bit different. So.

C

It's a charge, density sort.

A

Of like how much charges generated I think due to like a water flow, I, guess cuz like oh, like a.

G

Available right.

A

Right, that's probably right yeah exactly and what's like this, and then we know, if you don't have any fluid source. Divergence of U ship is zero. So, for instance like this, this earth dam problem, except for your boundary like you, is always continuous, so that like is usually zero in our interest of domain and like this, will be 0, so you're only getting like what what actually generates your signal is the difference of this QB in in the domain.

A

So, if you like, if your material have different QV in the domain, that this continuity actually generates, the SP signal so.

C

Qv, it's not a material, it's not material property. Yeah.

A

It's not not Cory I, it's like us. It's.

C

Like a run of fluids, right.

A

C

A

Yeah so that they're there needs to get like that. This continuity of that property, looking in Italy so L, is pretty similar to QP cuz I. Remember L was QB x, k, which is it's complicated, but at the end that the problem is pretty simple because now like here what your RM is actually all in the right hand, side, let's assume that Sigma is no okay, which is not quite true, but like. Let's assume that we got the DC array, we can get pretty good, continuous structure. So that's okay!

A

Then you need to choose the model like here. We have actually three choice. You can choose Q, which is like a divergence of j. Yes, so it's like it's like an.

C

A

Sex, basically a source, inversion problem and or you can invert for jas, which is a vector, yeah and or we can invert for H, which is I, believe head or also you can invert for velocity. Why lots of choices, but the here I'm considering two pieces where, which is that one is scalar q, + j s is the vector, so basically it's similar to mag inversion, if you actually like linear, mag, inversion or mvi version.

A

If you take an allergy, but I'm not using interval equation but PE to do that so yeah, that's actually the equation: a phi is equal to q and then, if you take the derivative, a is not depend upon M. So we don't have that term at this yarn term. But like we're getting this so depending upon what your choice of the model, this will be different.

A

So you simp AG and.

A

So here I'm showing the example. So this is a 3d model. That's a plan view at the like a 39 meter below, and then this is 45 meter below. So this is two depths lies, so we have a to leakage, paths and then that's a section view at here and then this one is here: okay, so we have a to leakage: paths which is like a two blocks: gonna in a not a half space with a dyke and then I define like 42.

A

So first I need to solve the flute like a Darcy problem, so I first simply define that usually at this and and that ends so.

H

Every just a little bit short.

A

Of like I, I'm not I'm not using yeah yeah I'm not like using up just like a tensor.

A

B

Demons imagine.

B

A

Very practical.

B

A

So solve like a Darcy equation, so here's our bomb, here's my boundary condition that I define and then here's the sort of fluid simulation resort, so basically a sec to solving DC problem but I with a different boundary condition. So you have one excitation.

A

So here we have high hydro, hydro, hydraulic, not hi, hi, Roley connectivity, oh yeah, like hydraulic conductivity at this leakage paths so fluid mostly flow here and then as a higher velocity, and we can see that so basically, it's have a conductive layer or conductive target, that's very similar, and then by using that, assuming you know the l or QV right. So you can compute the streaming current, which is like this.

A

C

A

Imagine how I don't like head looks like or hydraulic gradient. It.

C

Looks like so that's actually how.

A

The electrical fields looks like all the cars looks like so so I was asking. What is it like, so.

B

I'm showing the streaming con, where is the direction of.

A

B

Conduction of condoms.

A

C

Induction great movie, oz, yeah, yep.

A

It's kind of hard to see, but anyway, then, by taking divergence, you can compute the Q which is like charges. So you have a minus charges and plus charges here and then using that as a right-hand side, you can complete the synthetic SP data. Sp.

C

Data usually have a one reference.

A

Electrode and then just compute the difference. So it's like a poll poll response similar because I'm.

E

Not changing that reference.

A

Electrodes usually can.

C

I think of the conduction current as locate secondary due to those charges or a step backwards. Oh.

A

Well, like a conduction like a streaming, current is not it. I think it's still, but it's a right hand side. If you think about that's a right hand, side, and then that right hand side is generating. Actually, the conduction comes so all physical things. What you probably think it's it so.

C

I think the connection period is actually due to like those charges, building yeah.

A

Exactly but the streaming current is more like a fluid flow, so if you actually invert for like because direction of flow flow will be similar to it will be same not saying but like that streaming current will probably a good and it reflects how the fluid flow is happening in the earth. So by inverting that looking at that streaming, current yeah, you have a good idea of what is flow in the ground that that's.

G

Actually about.

A

Yeah yeah I think.

G

Use those charges you're exactly right right.

C

G

You climb from luck right.

G

Bobby there you get the positive charges on the right hand, basin yeah,.

B

A

Exactly that, so, actually, if you think about a little bit differently, if you think all of these on the surfaces and you're actually finding where your sources right like cuz, that's what usually happens all right and then you can even like think about, like the resolution, might be not very high for an in terms of depth, because your excitation is so simple compared like similar.

C

A

Gravity series are inversion site like this invert, this SP data first invert 4q. This is the like adepts lies and that's that that's cross sections, so I used to that's waiting, otherwise,.

C

A

Sensitivity is pretty high on the surface and this is the current density, so this is amplitude answer.

C

In this one, you have to.

A

So basically, this is true model and that's the that's a true okay and that's the recovered.

C

Ok, see you have sorry I thought I thought we had two different models, but this is actually just two different slices through exactly.

A

Yeah, so this is all same model, but the two different slices, because I I was curious, like oh can I get that information. Yeah cuz, like I, put a little bit different, dad yeah I think which is pretty hard target here. I think the difference was big, so I'm sort of like a little bit negative, not negative likelihood pessimistic research at that.

A

Well, I cannot distinguish this that that that's with a DSP would I who what I hoped was actually in the beginning by using them like that, the sparse sparsity constraint, I still pretty sure I can get much better reserve and I think I'm pretty sure I can get the right depth here, because I'm, if I'm pushing hard yeah it somehow like yeah that the performance is not great and in that and I ya f a couple ideas.

A

Why and then I think our optimization it a little bit like need to be a little bit mortified I guess current one is actually like a working for first-order problem for very smooth conversion, I think but d. But for your sparse inversion, I don't know I, not she wasn't sure here. You said you're doing some sort of that.

F

Whitley yeah they're also that.

A

There is also killing or.

F

A

Updating the depths waiting or you no, no, yes, I think they're a couple they're quite a couple like spices that I had to do, but in theory.

B

I fail, I will rock this in.

A

Theory should much you work much better and then you can probably like kind of get the good depth and same as currents so.

G

This is red, rose, yeah, yeah,.

A

This is the current density and amplitude it's a vector inversion. So if you look here, it's like like it's kind of hard to see, but it's a Vectra.

A

So it goes like this and it goes like that and that's a true and that's recovering, so it does.

C

The okay job, but.

A

It's felt so pretty smooth and like a getting a horizontal information is okay, yeah and then you may argue uh I just by looking at the data, I can still get the gorgeous location.

A

Sure I think this SP immersion needs more a confirmation like either you need to know the depth or you need to put some sparsity constraint other than that, like it's monitor, information that you can get from this inversions, not great yes, but I- think saw the chance, um probably a little bit more interesting thing is how I implemented, or are you working on? Oh this.

G

A

In the main, it's your personal yeah.

C

Yeah person point I the same.

A

Thing as pianist impact RC, but I did task, but but not seriously. Sigh like I just so here, I like I'm whole bunch of like the properties so type likes. I'm separate I'm, separate like, for instance, I'm when I'm inverting for vectors I separate for x, y&z direction and either like yeah qh.

A

What I hope I can get the property in the source, because that was my initial plan, but I cuz. All things actually happening in the source, so I hope that I can have a property in the source. I can pass the property map to the source and I can directly use their. That was what I hope that I don't know. What was the problem, but the there I have a little bit of problem, because I think that's probably cure arguments, because that the setting the keyword argument set the mapping I guess to rid of the.

H

Source doesn't inherit from the synthetic property puff hello.

A

H

It is triggering a mess up all right, yeah. I think I would write.

A

Yeah, so if you look at the how I implement it, it's that to basically what you need is eval, and then I have a three different eval and then above the rave is the sensitivity so yeah. If you want to invert for source, that's how you do it and then actually I am using, but probably using just.

A

Same j back in dec problem- I guess maybe oh maybe not.

C

A

I don't know actually maybe yeah yeah. I think I think im using actually same joy back cuz like it has all the things and yeah I. Don't need to do anything tonight because I had to do what I had to do.

B

A

Actually, developing this part and then plugging into the current DC problem and then actually sort of works, because the this one actually inherited this problem. That probably the reason why I did it yeah it's kind of cool like sometimes cuz yeah like if there's a pretty general chunk, then just need to inherit and then like. If that works for your Apollo, you can just use it ha cool.

D

Yeah, that's it.

A

Was it whatever see over yeah I think people were.

H

Crazy SP conversion syllabus leading practice there, as as the used for dem yeah.

A

That has been used for dad and iOS 8 by the data. No I think they didn't work. So I think this is sort of the cutting cutting edge like like that's what they do. So that's sort of a to kind of state of art and is what they like to blend right into the scalar.

D

Yeah yeah and what works comments cuz.

C

You know you had a chat with him. Well,.

A

If you think about I, think here, like if you're think about this right inside in working for Jas, is like kind of hard like is what you're right inside is divergence of j s, meaning like yeah, there's a null space that you cannot like recover because so actually q is about that's about the information than you can get, and that was his comment. Basically, so you can. You cannot get like no more information by using jes, but I don't know because I get.

A

That depends like upon like what your constraint of your inversion, because you can actually inject that information through your constraint. Then then, basically like no one is pretty similar, so I don't know, I wasn't like physically right but I, depending upon how you implement it's different and then especially current. If you limber for current it's more like a direct dog, that's how the current role, so that's I, thought that's! Okay, but having both is like I thought. That's better! It's like, depending upon.

E

A

Problem, your interest, you could choose differently and then I think as peace. Probably, why do we use because, like it'sit's, they are using for geothermal and also this can earn an example and most economy they are using for I can just get a little stream.

B

A

About that, oh- and there are quite a few different sources- there's a streaming like a full flow mate generates, and then chemical reaction also generates this yeah, sometimes using for mining application.

F

A

F

A

Right so actually me Sam he's is like visiting here and he was.

G

A

Like actually provided they, oh it did it Bertie yeah he's a mine.

B

A

Mike Epps, indeed, if someone has Judith and the problem so actually averted.

D

A

In that yeah we've recovered. Something like that and I. Think at one point, I showed you guys: oh yeah yeah, it's like it's.

B

Coming out of this.

A

Is set for it like coming like there were four coming from the earth. Hence we can be anywhere.

C

A

So there's a an end like dumb, that was his name Craig yeah.

B

F

Like yeah yeah.

A

Yeah, I think, for because he's problem is geothermal, so yeah he's got to eat it up. So that's a nice bit of potential yeah, hey! That's it! Yes, I hope. D I can test a little bit rigorously and then merge back to main branch where oh, it's not yeah. It's that it's local, its complete little more yeah, my like. If you look at the damn geophysics that what I was working with.

A

Oh yeah, it's a no! Oh this fields! Oh that's weird! hmm That's interesting, not sure what.

B

A

Not really hmm interesting, ah-huh I, don't know it's for a while, but the what I did was actually simply implementing kind of hair define.

B

The boundary condition and.

A

Solve bc problem that was it and.

B

A

A little bit of utilities that you can define like different boundary condition different sides that was dose it yeah yeah. I hope that I could have a little bit more time that I can implement like kind of non like a nonlinear I like connectivity, so I can like still state a state but like do sort of Jacobi iteration or something so that like not assuming but actually solve for the problem, but ya didn't happen.

A

Maybe at one point.

F

Ever get worse, there's nothin, gotm I could kill them.

F

A

I don't know, I think that really depends like it's. Like geothermal, it's sometimes couple kilometers.

F

A

I'm not sure, oh, how deep you can see with ESP, but I think that depends cuz like sometimes you have a pretty good fluid and that generates pretty high voltages, but the case so I it's one of our cube. So it's not going to be like you may not see one kilometer naps but I, don't know a couple hundred sure. Okay,.

A

But I think it's same same as gravity a mac, so I love them. You can see with the density because I sometimes x, people can see pretty dear, so I think that's the same. I guess same issue like how much how big scale you're talking you're sort of thing came. A big scale like fluid flow and that generates voltages probably can.

C

A

Deeper by Conger depp's as well: oh.

E

Oh yeah, good meat is good in terms of the receivers are having half that affect your.

E

Russian winter no depth ten to thirty well.

A

I said like a like a receiver geometry or age, yeah, yeah, yeah, yeah um well like I, think that, like that, doesn't help too much, because if you have a really fine scale features on on the surface or like that, then having a dense receiver agree that actually helps. But what you're, seeing from the deep like a I, don't deep at or whatever, if structure is pretty smooth signal. So I think you don't need that. Much like a receiver treat like a very fine receiver greet but yeah definitely helps.

A

But the increasing the receiver point doesn't help to get a best resolution in sp problem right. But if you have a like a large area that actually helps because it that sort of captures your like a full and one Amelie yeah yeah yeah I.

B

Think that's probably more.

H

Important, or are you going to be able to sample on the water side re? Are you technically only sampling on not.

A

Underwater oh, go for DC problem. Sometimes they get the data on the water zesty! That's not too hard because, like you put electrodes on the water yeah, we have a good contact so, but the internal like they're, usually do a permanent system for SP, so putting yeah I'm, not sure 11, yeah I haven't seen that I haven't seen like in.

C

A painful road pretty quickly right.

A

That's really yeah.

C

A

True, that is profound. Never on inheritance, you order yeah.

D

I think they usually put on the.

A

Crest side, which is not the water like this part, that's what they are usually do. I think clear.

B

A

Bro right right right what I hope they could do 3d not about a 2d not on one line. These couple lines then, like actually just seeing that data will tell you quite a bit of things otherwise, like.

C

A

No, it's it's not too hard. I guess.

C

Yeah well, I, guess um flossing once you just bring up was that dead? We request? Oh, yes, we can try to get in and just whatever comments that you're not to go too in depth just like, if there's things that changed, we've added, there's things that have been added to that would be great to have documented and then I think I know going forward. Maybe try to do this more so that we don't end up with the 13,000 minds, but I mean that's okay, it's great I.

H

Think we're a better, so yeah there's a vessel to it. There's a ton of stuff, that's changed and.

C

It was my work up. Yeah called anyone, be awesome to get this SP like image to pull request, as is whenever you have a chance, because I think that that could be really helpful, forgetting properties injected like into the survey, so they.

A

Haven't started.

C

Thinking about and burning for source, it's like in a natural way, restaurant.

A

Right yeah, it sounds like they got all the pieces and then a yeah. It shouldn't be our yeah I guess.

C

Oh, take us no thanks, ru thanks guys.

C

See kidney see.

E

You later thanks.