Material Uncertainty via Stochastic Analysis in Hyperstudy

Hi Everybody,

I am fairly new to Hyperstudy and I would like to use the Stochastic Analysis to investigate the scatter in the linear structural response of a loaded part due to material uncertainty.

I do not want to optimize anything.

Instead, I want to define material constants (i.e. E & Nu) with a mean and a normally distributed upper and lower bounds (e.g. +- 10%).

Then run the analysis and find out how much for instance the Maximum Stress, Strain or Displacement varies for the given load case.

I managed to define the model setup and the variables. The nominal run is complete.

Can I get uncertainty affected vonMises Stresses & Strains as well as Displacements (XYZ) and reacton forces on individual nodes?

These parameters are output variables in the Hypermesh analysis.

However, how can I define these response variables correctly in Hyperstudy?

I am using Hyperstudy 2019 and Optistruct as Solver.

Thanks for your help.

Regards, Sig

Find more posts tagged with

English

HyperStudy

Accepted answers

Diana_38240

suggestions:

HS-2000 - Step 3: Create and Evaluate output responses (e.g. Displacement on specific request (nodes)

HS-5020 - Stochastic study of a Cantilever I-beam

All comments

Adriano_Koga

this step of defining your outputs is done in the step 'responses'.

There you can use the 'file assistant' to select your nominal run H3D file and point some specific elements (take a look at some of the basic tutorials in help for more details on creating responses).

Once you have defined your responses (the metrics that you're going do measure) you can proceed to a stochastic study, defining your variables ans their distributions (normal, weibull, ...). Then you proceed to the specifications which will indicate your sampling method (LH, hammersley, mels, etc...).

The your model will be submitted, varying the inputs, and getting in the end the stochastic results from your responses.

I believe we have some tutorials for stochastics as well in the Help.

Diana_38240

suggestions:

HS-2000 - Step 3: Create and Evaluate output responses (e.g. Displacement on specific request (nodes)

HS-5020 - Stochastic study of a Cantilever I-beam

Sigfrid

Hello Everybody,

I recently came back to work on uncertainty due to material scatter again in static structural analysis.

The HM model is a three-point bending beam with enforced displacement (SPCD).

I would like to plot a load over dispacement curve including upper & lower bounds of the reactionforce for each time step.

I ran a stochastic analysis with material data (E, Nu, G) as variables.

For the response I defined the reacrion force at a specific node (SPCF).

In every run, the SPCF at the end of the analysis is computet and the final result is the distribution of this value.

Question:

How can I define everey increment during the loading of my beam as response (instead of the last/maximum value only)?

Regards,

Sigfrid

Adriano Koga_20259

Hello Everybody,

I recently came back to work on uncertainty due to material scatter again in static structural analysis.

The HM model is a three-point bending beam with enforced displacement (SPCD).

I would like to plot a load over dispacement curve including upper & lower bounds of the reactionforce for each time step.

I ran a stochastic analysis with material data (E, Nu, G) as variables.

For the response I defined the reacrion force at a specific node (SPCF).

In every run, the SPCF at the end of the analysis is computet and the final result is the distribution of this value.

Question:

How can I define everey increment during the loading of my beam as response (instead of the last/maximum value only)?

Regards,

Sigfrid

for plotting load x displacement you can use HyperGraph.

You can use the crossplot function as shown in this pdf:

https://insider.altairhyperworks.com/wp-content/uploads/2018/02/T-T-1292-HyperGraph-Cross-Plot.pdf

For your question:

what do you want to do with the intermediate values? only measure them?

The process is the same as creating the last/maximum.

HSt extracts a vector, in example, ds_1 {0 0.5 3.0 5.0 7.5}.

in your first model you probably used the max of the vector 'ds_1' as your response. Something like:

r_1 = max(ds_1)

You can just edit your response and ask for a particular index (the index is 0-based, so the first one is 0). Im pretty sure you use the [] to get a particular value.

r_2 = ds_1[0] >> this would give 0.0 from my example (1st index)

r_3 = ds_1[1] >> this would give 0.5 from my example (2nd index)

r_4 = ds_1[3] >> this would give 5.0 from my example (4th index)

Sigfrid

for plotting load x displacement you can use HyperGraph.

You can use the crossplot function as shown in this pdf:

https://insider.altairhyperworks.com/wp-content/uploads/2018/02/T-T-1292-HyperGraph-Cross-Plot.pdf

For your question:

what do you want to do with the intermediate values? only measure them?

The process is the same as creating the last/maximum.

HSt extracts a vector, in example, ds_1 {0 0.5 3.0 5.0 7.5}.

in your first model you probably used the max of the vector 'ds_1' as your response. Something like:

r_1 = max(ds_1)

You can just edit your response and ask for a particular index (the index is 0-based, so the first one is 0). Im pretty sure you use the [] to get a particular value.

r_2 = ds_1[0] >> this would give 0.0 from my example (1st index)

r_3 = ds_1[1] >> this would give 0.5 from my example (2nd index)

r_4 = ds_1[3] >> this would give 5.0 from my example (4th index)

Hi Adriano,

thanks for your quick reply!

"what do you want to do with the intermediate values? only measure them?"

I set up a model of a three point bending test in a structural simulation, where I looked at the reaction force acting on a specific node. Using the standard deterministic aproach I get a single load-displacement curve.

However, in my experimental tests the three point bending samples display some variability in the load displacement curves due to scatter in material properties.

Hence, I want to include this material uncertainty in my simulation using a stochastic analysis with E, Nu and G as variables. The goal here is to check if the experimentally determined load-displacement curves lie within the bounds of the predicted load displacement curve determined by the stochastic analysis.

In my first attempt, my output response was max(ds_1), which only gave me the maximum force (e.g. at maximum displacement). This is not what I want.

Now I tried to define the responses via "Multiple Items at Multiple Time Steps" & "Slice data source". I get a response for every increment of the loading process (as defined in NLPARM, NINC in my model). The analysis is currently running and I think it does what I want.

I will also try your aproach to extract specific results since computing every increment is very expensive...

Regards Sigfrid