Test data averaging is a new feature in the material calibration app in the 3DExperience platform as of R2024x FD01 (released on Feb 10, 2024 to the public cloud).  

The image above shows 7 repeats of a uniaxial tension test performed on 304 Stainless steel.  This raw data is attached to this post in a zip file below.  

Test data averaging has been requested by several customers.  I would like to make two comments:

1)  The best practice when dealing with several, or many, test repeats is to pick "the most representative test dataset".   There are a variety of reasons that averaging together test datasets is a bad idea, especially if you are simply averaging together all the noise in the data.  If you do average test data, you ought to first clean all the datasets.

2)  People imagine that averaging test data together is a simple task, often because they are thinking of just data.  In reality, we are really working with data, and averaging that sort of data can get complicated. 

Import test data as

In the video below, we show the import of test dataset #1, and all of the other imports are similar.  In the attached zip file there is a 3dxml file, named 304_data_7repeats_import_strain_stress.3dxml, that just contains the 7 datasets.  For this section, we import just data.  This is the simplest use case.  One of the simplifications is that the strain-rate for each dataset is then identical.  

After importing all of the 7 datasets, my calibration object looks like this:
 

The data looks like this:
 

Zooming in to the linear range:
 

From the two images above, I can already begin assessing the quality of the 7 tests. All of the tests have some zero-shift issue, which we can correct in the app.  In the testing of metals we expect a pretty tight grouping of data in the linear elastic region, from image #2 we could conclude that test #1 and test #2 are of poorest quality and we should not use those tests.  We have a new feature that can "deactivate" a dataset.  After deactivating test 1 & 2, our test data looks like this:
 

Tests 3 & 4 are very close to each other.  Tests 6 & 7 are very close to each other and test #5 is in between. After some further inspection, I chose to keep tests 3, 5 and 7.  Prior to averaging these 3 datasets, perform a zero-shift on each one.  The video below starts by using the 3dxml file named 304_data_7repeats_import_strain_stress_cleaned.3dxml

Import test data as

The import of the data is similar as before, except this time we select all 3 columns of test data .  This injects a bit more variability into the data.  The plot below shows the nominal strain versus time. The strain-rate for each dataset is no longer the same.  For metals, we do not expect that this small variation in strain-rate should effect the response physics, nor do we need to capture strain-rate as a material model parameter. 
 

In the attached zip file there is a 3dxml file, named 304_data_7repeats_import_time_strain_stress.3dxml, that just contains the 7 datasets.

There is also a file in which datasets 3, 5, & 7 have been cleaned (zero-shifted):
 

304_data_7repeats_import_time_strain_stress_cleaned.3dxml

The zip below contains these 5 files:

304_Data_7Repeats.xlsx

304_data_7repeats_import_strain_stress.3dxml

304_data_7repeats_import_strain_stress_cleaned.3dxml

304_data_7repeats_import_time_strain_stress.3dxml

304_data_7repeats_import_time_strain_stress_cleaned.3dxml

Back to:  Sharing Material Test Data

Back to:  Material Modeling and Calibration - An Overview and Curriculum