you are here:   / News & Insights / Engineering Advantage Blog / Tips & Tricks for Matching FEA Results to Physical Data

Engineering Advantage

Tips & Tricks for Matching FEA Results to Physical Data

Preferred Location for Strain Gauge | FEA Consultant
July 18, 2017 By: Peter Barrett

Confirming finite element modeling assumptions with test data, when possible, is the ultimate validation tool. This blog provides some tips and tricks to effectively and efficiently correlate analysis results with experiments.

In my first job out of college over 30 years ago, I was involved in both pre-test and post-test predictions of the structural response of ship hulls subjected to blast loading. The test fixture was basically a stiffened barrel that represented a ring stiffened hull.  It was extensively instrumented with accelerometers and strain gauges for correlation between test and analysis. This “barrel” was tested in the open ocean waters where a large TNT charge was ignited in close proximity, such that significant damage occurred to the “hull” and its internal stiffeners. Our team performed time history structural dynamic finite element simulations prior to the test. We published, along with our competition, our predicted dynamic response, including displacements, stress and strain histories. In addition to determining the true strength of our modeling prowess with these blind predictions, the pre-test analyses from all participants were used to define the critical locations to be instrumented during the testing. Post-test engineering consisted of comparing test data with the pre-test simulations followed by re-running the analysis models with modifications to assumptions predicated from lessons learned from the test. From these particular physical tests, we learned that the ability to accurately simulate the local stiffener buckling was a key to matching the experimental tests. By improving the analysis model to better simulate this part of the test data, dramatic improvements were made to the correlation between the test and analysis data.

Fast forward more than 30 years, and the challenge to correlate test and analysis data still plays a major role in the engineering simulations of complex structures.   Some of the key lessons that I have learned to best correlate between test and analysis over the years are:

1) Placement of Test Gauges are Critical

(a) Accelerometers should be placed where deflection are large, but be careful that their weight and connectors do not influence the model mass and stiffness.  If this is a concern and there is no easy workaround, then these gauges can always be added to the simulation model as a combination of a lumped mass and spring.

(b) Strain gauges should be placed in areas of high stress, but away from singular stress regions. A rough rule of thumb is placement in areas where one’s stress contours are large, but spread out over a significant portion of the part surface (See Figure 1, above).

(c) Pre-test simulations can be used to determine these areas of uniformly high stress while avoiding areas of steep gradient.  When gauges are in areas of large gradients, a slight mismatch in position can significantly impact results variability and the ability to match test data.

2) Make Sure the Test Loading is a Measurable Value

(a) In FEA, the magnitude of load only matters when you have significant nonlinearities.  For testing setup, load magnitude can have a significant impact.

(b) For example, take a typical load cell that might have 0.1% accuracy, so a 10,000 lb. capacity load cell is accurate to +/- 10 lb.  Therefore, using this load cell to measure a two or three-pound test will not be reliable.

3) Accurate Boundary Conditions are Key

(a) For example, a fixed vs. pinned support is purely a modeling assumption.

(b) Running sensitivity analyses on a simplified model is an easy way to determine the impact of the FEA support assumption.

(c) If the variation between the fixed and pinned supports is unacceptable,  then one can add additional simple features to the FEA model.   An elastic foundation, linear or nonlinear springs and/or beam elements can be used to enhance the accuracy of the model support infrastructure.

4) Extract Consistent Stress and Strain Results to Compare with Test Data

(a) The most common test data consists of surface strains with a specific orientation extracted via a strain gauge.

(b) In ANSYS, one can layer shell elements on the surfaces of solid elements in the gauged areas and then extract surface stresses directly from these shell elements where the element coordinate system is aligned directly with the gauge. Thus, a single shell can be placed for each strain gauge where this dummy shell adds no stiffness to the model but allows for the generation of correctly oriented surface stress and strain data that can be compared directly to tests.

5) Start Simple with the Simulation Models

(a) For material model validation, model the actual test specimen.

i. In characterizing small polymer samples, for example, I have used FEA simulations to help define the radii on the dog bone sample used in the physical testing.

(b) When matching forensic data, start with local models and then build knowledge needed to create the global models, using as many simplified assumptions as possible.

i. Investigating the thermal-structural response of the collapse of the World Trade Center was the most complex simulation I’ve ever worked on.  We started with simple models of the critical connections such as the truss-to floor and truss-to-wall supports.  Based on these simple models, we created single “break” elements that would “fail” as a function of force and temperature to accurately represent connections breaking as verified by the simplified models.  Figure 2 illustrates the implementation of these “break” elements (for details see

Figure 2: Example Summary of "Break" Element Analysis Results

6) Agree on what Constitutes a Tolerable Level of Calibration Before any Simulations are Performed

(a) Acceptable error is very problem dependent.

i. For a linear elastic small deflection analysis of a metallic structure one would expect the error to be less than 1%

ii.However, for a very complex assembly of many parts, error in the 20% level may be the best one can achieve.

(b) It is important to use engineering judgement here.  A simple analogy I like to think about is that the problem you are solving can be visualized as either a standard or inverted funnel (See Figure 3).  Your objective is to take the known data funnel the unknowns to get a calibrated solution within acceptable engineering accuracy while isolating the sensitively of the key input parameters.  Resist the urge of inverting the funnel where more questions are created than answered with each addition to the FEA model.  With the standard funnel strategy, you minimize the number and size of the simulations needed to gain accuracy consistent with your engineering goals.  Getting relative data on multiple input parameters is much more valuable than a single accurate solution with limited understanding of the impact of the different modeling assumptions.

Figure 3

What tips and tricks are in your tool box that you have applied to simulations to match test data?  I would love to hear from others about their real-world experiences.