Inventor of medical devices, the man behind the Segway, and FIRST backer Dean Kamenvisited the Phoenix area yesterday and today and we were lucky enough to have people from PADT invited to two different events at which he spoke.  For engineers involved in product development, this is like a visit from an NFL quarterback for most people.  He turned out to be open, engaging, and a very good speaker. 

We could go on in adoration and explore the guilt and envy we feel after seeing all that he has done.  Instead we thought we would highlight two things we learned from his visit:

1. The FIRST program that he started and still heads is making a huge difference in this country and around the world.  PADT has been peripherally involved, focusing instead more on the underwater robot scholastic competitions that are very popular here in Phoenix.  But FIRST is now huge, and is still growing.  But what we learned is the positive impact it is having: Students who participate in FIRST are 3 times more likely to become engineers, 30% more likely to attend college, and twice as likely to volunteer in their communities. Those are some positive numbers.  Those of us in the engineering world should take advantage of that and support FIRST.
   
                                                                     http://www.usfirst.org/
   
2. Second, he offered a unique perspective on how engineers see the world.  When he was young he heard the story of David and Goliath.   Most people see a religious message in this story, there are various interpretations. But as a child, Dean Kamen did not see those messages.  What he saw was that David won because he had better technology. He had a sling shot.  That is how he beat the giant.  I found that a very interesting point of view. If you don’t get it, ask an engineer.

If you ever have the chance to explore what his company, DEKA, is currently doing with a revolutionary power generation and water purification solution for areas of the globe without power or clean water, do so.  It is very leading edge stirling engine and distillation technology. 

This weeks Focus posting is not going to be very technical. In fact, it is a bit of an editorial.

Over the past five years or so we have seen a lot of companies who use ANSYS, Inc products move away from traditional face-to-face training with instructors in a classroom.  There are a lot of reasons for this.  The two most common are that 1) the company does not have a travel budget and 2) that training labor hours are considered overhead and managers all have very strict overhead restrictions.  On top of these two, many companies are just plain trying to save money on the cost of training or limiting their overall training budgets.

What we have seen is a larger number of users either trying to train themselves from manuals or downloaded training material, or people trying to do web-based training.  One can certainly learn how to use an FEA or CFD tool this way, but through our tech support we are starting to see the negative side of this shift: users only understand some of the aspects of the tools and do not have a depth of knowledge that goes beyond the basics.  So when they run into a problem that requires a move beyond those basics, or that might require a more nuanced approach, they struggle or they call tech support for on-the-spot training.

Interaction

Even though we engineers are not the most social sub-species of humans, we can still heavily benefit from face-to-face interaction during training.  When PADT teaches a training class we find that a small portion of the time is spent lecturing on and doing workshops for the basics.  Most of the time is spent answering questions that occur to students while they take these basics in.  Some are industry or user specific, some delve deeper into the tool than the training material does.  But they all provide an education to the whole class that never occurs otherwise.

We have taught, and been students in, web based training classes. The interaction is just not the same.  There are not as many questions and the instructor is not able to use body language clues to see if the class is really getting what they are saying.  In fact, we feel this is the biggest issue. When you are on the phone and sharing a screen you can not even tell if the students are listening.  So the instructor pushes on, the students drift further away, and the true benefits of the class are lost.

Make a Case for Classroom Training

The point of all of this is that we feel users out there need to make a case for real classroom training.  When your boss says that there is no travel budget, not enough overhead allocation, or just not enough money, argue strongly that the cost differences of online or self training are not that significant when compared to the productivity problem of not having deep, interactive training.  If you are a boss, admit it, you know we are right.  You should fight a bit harder for the budget because in the long run you will save money.

Another way to look at it is the relative cost of classroom training versus how much you will use the ANSYS tools you are trained on.  Even if we assume that the company you work for kind of sucks and most engineers move out of there in five years, one to three week of training is nothing when compared to five years as a user.  If your productivity is just 5% higher during that time, the savings are significant. 

Do the full classroom training.  You will not regret it.

As a full disclosure:
We are partly motivated to express this opinion by the fact that we make money doing such training classes, but in reality very few of you reading this will do training with us (although you could use us if you wanted to… hint, hint).  Most of you do your training through other Channel Partners or ANSYS, Inc.  So this posting is not entirely self serving.

About the pictures:
I find the stock photographs of what are basically models so contrived and stereotypical that they are hilarious.  So I grabbed few of my favorites to share.  I love it when they always have someone crossing their arms, looking thoughtful. 

Okay, so it’s not the ugly critter in the picture, (that on was found at my house), but the bug in ANSYS R14 can give you the willies just the same. ANSYS, Inc. is working very hard to get the situation remedied, and they will be sending out defect notices shortly.  They provided a quick-fix however, and we wanted to get it out to you as  quickly as possible. So I have done some screen captures while I ran through the fix.

The issue found is that if you close a Mechanical, or Meshing, session and then hit ‘Save’ on the project window, you may lose the contents of your Mechanical database.   The problem has arisen in R14 because ANSYS, Inc. has decreased the start up time of Mechanical by pre-loading it when Workbench is started. When you close Mechanical, it stays open in the background as an empty database. There is a 1-second ‘window of opportunity’ after you close the Mechanical editor and save the project file when the process threads are not fully synchronized.  If you save during this period, the blank session gets saved on top of the good session, and all data is lost.  If you wait longer than a second, there shouldn’t be an issue, but customers have been reporting longer times where they have still lost data.  ANSYS, Inc. is working with those customers to find out what is causing the longer times on their boxes.

Luckily the remedy is simple, and hopefully none of you will have to lose any data.  Since the issue is caused by the pre-loading of Mechanical, the remedy is to simply turn the pre-loading off.  This has to be done from  the Tools > Options menu on the Project window. Just uncheck the Pre-Load box in both the Mechanical and Meshing dialog boxes, and then close Workbench after hitting ‘OK’. The next time you open Workbench, the bug will be neutralized.

I just wish it was that easy for that creepy guy in the picture!

One of the most common questions on XANSYS, and a common tech support question for us is: “Is there any way to make my result file smaller?”  In fact, we just got a support call last week on that very topic.  So we thought it would be a good topic for The Focus, and besides the standard quick answer, we can go a bit deeper and explain why.

Why so Big?

The first thing you need to understand is why the result file is so big. One of the fundamental strengths of the ANSYS solver, back to the days when John Swanson and his team were writing the early versions, is that the wealth of information available from an FEA solution is not hidden from the user.  They took that approach that whatever can be calculated will be calculated so the user can get to it without having to solve again.  Most other tools out there do the most common calculations, average them, and then store that in the results file. 

If you go back to your FEA fundamentals and take a look at our sample 2D mesh, you will quickly get an Idea of what is going on.

When an FEA mesh is solved, the program calculates the Degree of Freedom results (DOF) at each integration point in each element. For most meshes that is at the corner of each element.  This DOF solution is then used to calculate the derived values, usually stress and strain, for each node based upon the solution at the integration points for each element. Then, during post processing, the user decides how they want to average those element

So the stress in the X direction for Node 2 in our example is different for element 1 and element 2.  Element 1 uses the results it calculated with the elements shape function to get stresses and strains for node 2, and element 2 does the same thing. 

Now most programs, they average these value at solve time and either store an element average, or a nodal average.  But ANSYS does not.  It stores the values for each node on each element in its results file.

By default, node 5 in our example will be stored four times in the ANSYS RST file, and each instance will contain 12 result items for a linear run (SX, SY, SXY, S1, S2, S3, SINT, SEQV, EPELX, EPELY, EPELXY, EPELEQV) and even more for 3D, non-linear, and special capability elements.

Storing all of this information is the safest option. Everything a user might need during post processing is stored so that they do not have to rerun if they realize they need some other bit of information. But, as is usual with MAPDL, the user is given the option to change away from those defaults, and only store what they want. OUTRES really is your friend.

You Are in Control: OUTRES

OUTRES is a unique command.  It is cumulative.  Every time you issue a command it adds or removes what you specify from what is stored in the RST file.  The basics of the command are shown in the table below, and more info can be found in the online help.  Use OUTRES, STAT to see what the current state is at any point.  Always start with OUTRES,  ERASE to make sure you have erased any previous settings, including the defaults.  

Remember that this command affects the current load step. The settings get written to the load step file when you do an LSWRITE.  So if you have multiple load steps and you want the same for each, set it on the first one and it will stay there for all the others.  But if you want to change it for a given load step, you can.

If you are a GUI person you can access this from four different branches in the menus:

Main Menu>Preprocessor>Loads>Analysis Type>Sol'n Controls>Basic
Main Menu>Preprocessor>Loads>Load Step Opts>Output Ctrls>DB/Results File
Main Menu>Solution>Analysis Type>Sol'n Controls>Basic
Main Menu>Solution>Load Step Opts>Output Ctrls>DB/Results File

The first thing to play with on the command is the first argument: Item. 

The default is to write everything.  What we recommend is that for most runs, using OUTRES, BASIC is good enough.  It tells the program to store displacements, reaction forces, nodal loads, and element stresses. The big thing it skips are the strains. Unless you are looking at strains, why store them.  Same with the MISC values. Most users don’t ever access these. 

The next thing you can use to reduce file size is to not store the results on every element.  Use the Cname argument to specify a component you want results for.  Maybe you have a huge model but you really care about the stress over time at a few key locations.  So use a node and element components to specify what results you want for which components.  Note, you can’t use ALL, BASIC or RSOL for this option.  You need to specify a specific type of result for each component.  Remember, the command is cumulative so use a series of OUTRES commands to control this.

Use What Works for You

The help on the OUTRES command has a nice example where the user specifies different solutions for different sub steps. Check it out to get your head around what is happening:  // Command Reference // XVI. O Commands // OUTRES

Next time you run a small but typical model for you, play with the options.  Most of the time when I have a lot of load steps or a large model, I use the following in my input deck:

OUTRES,ERASE
OUTRES,BASIC

Sometimes I only care about surface stresses so they use (of course last can be replaced with all or any other frequency):

NSEL,s,ext
ESLN,S,0
CM,nxtrnl,node
CM,extrnl,elem
OUTRES,erase
OUTRES,nsol,last,nxtrnl
OUTRES,strs,last,extrnl
OUTRES,nload,last,extrnl
OUTRES,stat

Anyone who has had to mesh shell assemblies has probably run into trouble with edges that don’t quite line up, edges that meet in the middle of faces and other problems that make the meshing process difficult.  Often geometry operations were required to reconcile those problems and many times significant effort was required to get a continuous mesh.

Another historically used tool to connect shell assemblies was the use of constraint equations to connect edge nodes to surface nodes on the finite element level.  More recently, advances in contact technology have allowed for the use of nonlinear contact elements to connect shell assembly meshes.  Both of those techniques, while useful, have some drawbacks.  For example, constraint equations do not support large rotations of the geometry as the direction of application does not change as nodes rotate.  Also, contact elements increase the computational expense if they can otherwise be eliminated.

ANSYS, Inc. now provides us with another technique for handling shell assembly meshing, called Mesh Connections.  First available in version 13.0 and enhanced in version 14.0, mesh connections use the mesher itself to connect shell assemblies toward the goal of getting a continuous or conformal mesh across the surface bodies that make up the assembly.

Consider this boat hull example.  It consists of panel surfaces defining the hull as well as some stiffening ribs.  All geometry is composed of surface bodies. 

Some of the ribs line up with edges in the hull surfaces, while others do not as shown in the close up image shown below.

We can now create mesh connections in the Connections branch after loading this geometry into the Mechanical editor in Workbench 14.0.

Upon generating the mesh, the mesher will attempt to create a continuous or conformal mesh even though we have do not have intersecting geometry. 

With the default settings, we can see in this image that it did a fairly good job of creating the mesh for the ribs which do not intersect with the hull surfaces.  Nodes on the hull surface were adjusted so that they connect to the rib geometry. 

In this case with relatively little effort we were able to obtain a continuous mesh between the ribs and the hull, even though the several of the ribs shared no intersections with the hull surfaces.  In fact, the mesh connections were able to overcome small gaps in between the geometry as well.

In 14.0, the mesh connections are generally performed after the initial mesh is created by default.  This means that if changes are made only to mesh connection settings, the remeshing operation is fairly quick since the initial mesh does not need to be regenerated in most cases.

Note that mesh connections exist in the Connections branch, not the Mesh branch. The mesh connection setup works in similar fashion to contact region creation in that searching for edges/faces to connect is based on proximity. The proximity value can be controlled via a slider or by entering an explicit distance, both available in the Mesh Connections details window.

To activate mesh connections, highlight the Connections branch and click on the Connection Group button in the context sensitive menu above the outline tree.  Change the Connection Type to Mesh Connection in the details.

Next right click on the Connections branch and select Create Automatic Connections.  You may need to adjust the auto detection tolerance in the details to make sure the tolerance distance is large enough to detect desired gaps between edges and faces or edges and edges for the mesh connection to work.

If any contact regions have been automatically created that you want to replace with mesh connections, delete or suppress them.  You have the choice of automatically creating mesh connections or manually creating them.  Both options are available by right clicking. 

In the example shown here, mesh connections are edge to face.  Edge to edge mesh connections are also available.

With a couple of mesh settings added, we can obtain a better mesh:

Note that the hull surface nodes have moved a bit in order to allow for the mesh connections with the ribs.  Here is a view of the outer hull surface in the mesh connection region:

There are other considerations as well, such as which geometric entities should be the master or slave.  In general slave geometry is ‘pinched’ into the master geometry.  Also, mesh connections can be setup manually for cases where the auto detection is not appropriate or is not providing the desired level of control.  Note that the mesh can end up as an approximation of the geometry since the mesh will have moved to close gaps.  Here is an example:

In summary, mesh connections are another tool that are available to us in ANSYS meshing capabilities, having value for shell assemblies.  In cases where shell geometry edges do not meet at intersections we can still obtain a continuous mesh without having to perform additional geometry operations.  Mesh connections can be faster than using contact elements at the edges as well.  There are other features and considerations for mesh connections which are explained in the ANSYS 14.0 Help.  We recommend you give them a try if you are tasked with simulating shell type structures.