Dear all, Do any body have ideas about EFD software for thermal simulation? Right now i am using Flotherm software for thermal simulation of power supplies, adaptors and chargers. Now i am planning to use EFD. Which one is better in terms of accuracy. What are the advantages of EFD over Flotherm? which one would be better in this field. any body have ideas about the pros and cons of both.

your invaluable reply can help me to choose the software in better way.

thanks Jitender

I think to use EFD for thermal calculation will be a advantage for your future work. Flotherm is more suitable for PCB-calculation because PCB have rectangular shape. For complex geometry I recommend other CFD codes like EFD, Ansys CFX or StarCD.

About accuracy of EFD I can't make a commend because I haven't used EFD. But to simulate with Flotherm you have to reduce details of your model. If your model is very complex, so it is very difficult to have the same volume like the original geometry. Accuracy is depending on the thermal volume of your model etc.

EFD has now a electronic module, so you can easy add assemblies like fan or heat sink in your model like in Flotherm.

Right now, I am also looking for other code to make thermal simulation for electronic cooling, too. My experiences with Flotherm is good, but due the complexity of model of our products I have to find alternatives.

I think Flotherm is out of date for thermal simulation, because you can't handle complex geometry with this software. The accuracy of Flotherm, as you have described http://www.cfd-online.com/Forum/main.cgi?read=58880 , isn't good enough. The sole advantage of Flotherm is easy to use. But what bring me easy to use when the result doesn't accurate.

Have you make a benchmark about CFD-software before? I think Ansys CFX is a good alternative for thermal simulation. Nowadays more and more engineers choose CFX for thermal simulation in electronic cooling field. With the Ansys Workbence, it software is now easy to use.

EFD is also not bad but cause of the grid technology you will haven't operation interface with your customers. I mean when you are using EFD and your customer is using StarCD, you haven't chance to change thermal model. But when you are using CFX, you can change the grid with your custumer because CFX, StarCD, Fluent, and more are using the same grid technology.

Thomas

The majority of electronic thermal models are characterised by a high level of clutter of predominantly cartesian type geometry. By clutter I mean the existence of 1000s of parts with 1000s of solid-solid and solid-fluid interfaces. It is exactly this type of geometry that Flotherm excels at and is by far and away the mainstay of its application. Glib comments about it's lack of applicability are somewhat frustrating but come as no surprise.

As one who has done a lot of work with Floworks (same thing as EFD.lab or near enough) in electronics applications though I wouldn't call it detailed board level stuff though component temperatures were the goal. I would call these more instrument applications - lidars, projectors, laptops, PC's and other medium power applications (say in the couple hundred watt to norht of 2 Kw of heat rejection plus or minus with considerable geometric complexity) where convective cooling is the primary uncertainty. Floworks/EFD.Lab is easy to use, gets answers that are quite close - say between a couple and 5 degrees on easy to measure surface temperatures.

Haven't used flowtherm but I really doubt, given what I know about it which isn't much, that you can turn around design variations of conserderably different geometry as easy as you can do it in Floworks. it works great for me though it doesn't support putting in the in-chip resistances explicitly as flowtherm would. My view is that if you are relying on CFD to cut very tight margins you need to be doing a lot of work on very similar problems with alot of experimental validation and model tuning. Floworks doesn't have much problem easily getting it in the ball park as mentioned on any suitable problem you through at it, at least for me and many people I know that use it. Many "conventional" cfd analysts disparage the code with very little or no experience with it. The Flomerics guys paid a very steep price to acquire the technology so it stands to reasons that they have some clue about what they bought into. My suspicion is that they will migrate all their CFD tools to the EFD technology platform over time as they refine its applicability and figure out how they migrate their customer base of flowtherm users with out ticking them off. Is it perfect? No but it is really very good for the right type of work. I have seen it produce nearly identical results to Fluent (actually closer to the experimental values but not very significantly) on exactly the same problem. As one analyst put it to me "I just don't get how Floworks can produce the same answers as Fluent with 200k cells and 100 iterations in a couple hours as fluent does with 500k cells in 1800 interations and 8 hours". The stuff is pretty good but hey it depends on how close you want to skin that cat.

I am using Flotherm for a long time and it is quite good for forced convection, but moment it comes to natural convection, it lacks in accuracy. Do any body using EFD this time and what is its accuracy.I think main parameter in selecting the software is its accuracy. if thermal simulation result is accurate at its best then electrcal and mechanical designer can take direct action based on the thermal result. But if it is not accurate, designer will be afraid of taking decision based on the simulation result. thus, i believe accuracy is the first thing to consider. So do any body have idea about accuracy of EFD.

To generate accuracy result with thermal simulation, you have to consider a lot of parameters. 1) physical models like turbulent modeling, Renold's Average Navia-Stokes (RANS), Direct Numerical Simulation (DNS) or Large Eddy Simulation (LES) 2) Quality of cells 3) Boundary Conditions

You have to check these things before you buy a code.

Hi Jitender, I find this interesting, and I think its a comment that's been made before. Going back many years I would say that the accuracy with natural convection is generally better than for forced convection. The reasons are severalfold: often users does not model fans with sufficient physical detail (hub, swirl etc.), the fan curve imposed was measured under different conditions with unrestricted flow upstream and downstream of the fan whereas the application has obstructions, use of a plenum where the pressure plate is not well characterized, forgetting to switch on the effects of gravity, etc. Also NC systems are self adjusting - if the temperature is too high, the flow will also be too high, which will reduce the temperature. Whether its natural or forced convection, the code is solving the same equations, so I would look critically at the grid adjacent to the board, around components etc. in the context of the results you've got and ask yourself whether you've got enough grid to resolve the gradients adequately - and don't forget radiation. Very important in NC. Again, look at the temperature gradients and decide how accurately to subdivide the key heat transfer surfaces.

Hi Bill, I think its a case of horses for courses. For complex non-Cartesian geometry EFD has a lot to recommend it, and being CAD embedded it means you can change the geometry within the design flow, so there is no 'staleness' problem of working on old designs (aka versionitis). FLOTHERM uses a Cartesian grid. Up to a point it can handle non-Cartesian geometry with objects for prisms, sloping blocks, etc. but its real power comes from the fact that it deals with a Cartesian grid. On that grid you can have locally Cartesian grids attached to objects with an N:1 interface at the boundary, so very rapid grid refinement is possible around components. These grids are attached to the object, so if the object moves, so does the grid plus all the other grid adjusts to accomodate this. This makes the entire system extemely (not a word I use often) robust, to the extent that we can fully automate moving objects within the grid, with collision detection. This is done in the Command Center - a utility used to drive FLOTHERM in this mode which has DoE and optimisation routines built in, so for example a heat sink can be designed to minmize its weight or pressure drop whilst keeping the component it sits on below a target temperature. In electronics cooling design times are very short and the users very often have very little background in CFD, and for these users, and particularly those in telecoms, networking and computing sectors, FLOTHERM is by far the best choice.

The Floworks /EFD products seem to produce good results in natural convection problems in my experience. Once in a while erroneuos results have been observed but there are really obvious and increasing cell count typically fixes the issue.

Thanks for the insight. Interesting. The optimization capabilites would be a great addition to Floworks & EFD users alike. Given EFD's primarily a cart mesh of sorts it would seem fairly straight forward to implement a similar optimization scheme. DoE gets pretty expensive with more than a couple of variables in my view as far as compute times are concerned though. Then again, the batch solve in Floworks/EFD lab on say and 8 processor box with appropriate/sufficient RAM would seem to be a force multiplier on this front for DoE optimization.....running 8 experiments in the time it takes do 1. Overnight optimization of at least 5 or so variables would be in the realm of possibility for say a few hundred thousand cell models. Thoughts? What would a comparable solve time for Flotherm be for similar optimization problem?

My comments on Flotherm were really aimed at "effort hours to first answer" sort of productivity. It is my understanding and feel free to straighten me out if I am out to lunch here, is that the analyst typically tends to pretty much build up models from scratch in Flotherm. Is that in fact correct or can one import and easily use CAD geometry and/or IDF files? It seems hard to believe that Flotherm would not have robust interfaces to (E)CAD systems as building models from scratch is very time intensive and not much fun regardless of the specific tools employed - just the tracking down of measurements for placing things and then entering the data is quite time consuming and tedious.

Hi Bill, I would say a 500K cell problem should solve in 1-2 hours in FLOTHERM on a single processor 1Ghz laptop. This is likely to be a lot faster than any unstructured code. With the Command Center we can run simultaneous jobs from a DoE matrix on different machines and it collects the results and does the optimisation.

FLOTHERM is used both for final design verification with lots of detail, etc. and at the other end for concept design. FLO/PCB is used this way to. Both have sophisticated data import from both the EDA (some native formats supported, others via IDF) and mechanical CAD worlds (IGES, SAT, etc), with chip package models etc. coming from FLOPACK, etc. Geometry can be created very easily in FLOTHERM using the Drawing Board which as the name suggests works like a very simple CAD tool for creating geometry with click to create and then graphics stretch to resize, move, snap to object, etc. It can also import power lists, etc. so it has a lot of sophistication that's focussed on electronics applications. A little off topic, but hopefully that helps.

Oh, and I forgot to mention that just because its natural convection doesn't mean its laminar - another mistake that users sometimes make.

Thanks for the info. You guys are obviously in the best place to know the subtleties on the various codes applicability.

Any chance you get the DoE optimization on the EFD/Floworks enhancement list? It would still be pretty handy and a little beyond what's currently available.

hi

flowtherm is specially designed for electronic cooling.

with EFD u can reduce the time for fluid domain extraction and meshing. even for complex geo s also it works. but EFD does not have options like other cfd codes to change the differencing schemes, under relaxation params. every thing is set inside by default. may be useful who have less interface with CFD. but for specilly thermal applications Flowtherm is good.

I would agree for complex problems going to codes like CFX, Fluent or StarCD. They are great general purpose tools. I wouldn't recommend EFD for anything but the simplest order of magnitude modeling. Flotherm as well is clunky. Try Icepak for your electronics work. much more robust solver than EFD or Flotherm.

good luck