Heavily relied on in the product development phase, experimental testing has historically helped design engineers evaluate component performance and estimate a product’s functional life. As priorities have shifted focus to more cost effective solutions and reducing time to market, industries are increasingly looking at Computer Aided Engineering (CAE) tools to meet design targets without compromising on customer expectations. The introduction of more robust computational capabilities in CAE tools over the last few decades have enabled engineers to use CAE to perform complex analysis and compare predictions with test results to strengthen their confidence in these simulated results. This evolving technology has led engineers as well as managers to challenge how they can better correlate Finite Element predictions (FEA) and test results to minimize the need to run costly repeated experiments.
Historically the static FE solutions have relied upon the worst case loading scenarios (say 90th or 95th percentile loads) as inputs and their results compared against static montonic properties to determine the success or failure of the design. However, with improved FE capabilities, there is a growing need to generate more accurate FEA predictions to capture dynamics effects, fatigue issues etc. that are crucial to suggesting right design changes. In order to verify these design changes, it is imperative to correlate FE predictions with newly measured test results or already available historical data/guidelines. This has led to the introduction of many “correlation tools” across different loading domains. HBM Prenscia has delivered a number of correlation tools for increasing the confidence in the results of simulations in nCode DesignLife over several releases including Virtual Strain Gauge, Virtual Sensor, Strain Gage positioning and Loads Reconstruction.
This article describes two of these correlation tools available in nCode DesignLife Base with some notes on user best practices and the value of using such tools.
Virtual Strain Gauge allows users to position “virtual” strain gages on the FE model in user defined reference directions and correlate the virtual strain histories with actual test data to gain confidence in the FE model predictions.
The Virtual Strain Gauge function highlights two powerful nCode DesignLife features; handling multiaxial load cases and the ability to position virtual gauges at specific locations and orientations to extract virtual strain histories.
In this example, we will consider the trailing arm of an automobile suspension that has been solved for unit load cases about a central load point – X, Y, Z normal loads and three moments (Mx, My and Mz) with the appropriate boundary conditions. We also have actual measurements from the wheel force transducers that tell us how each of these six load cases evolve as a function of time. The physical part was tested in service with a uniaxial strain gauge installed in an area of interest. Now nCode DesignLife's Virtual Strain Gauge will be used to predict the strains in this same location.
As seen in Figure 1, the FE model and actual measured data form inputs to the Virtual Strain Glyph. At every node in the FE model, this tool scales the stress state due to every unit load with its individual scale histories (time history loading). If all six load cases are used together, then it does a linear superposition of all the six stress time histories to generate a combined stress history at every node throughout the model. Since the actual strain gage location installed in the test is known apriori, a Virtual Strain Gauge is positioned at a node closest to actual location in the FE model. Now at this node of interest, the Virtual Strain Gauge Glyph automatically exports a combined test history that can be compared against actual test results.
The XYDisplay glyph now shows two channels of time history data: the virtual strain (in red) and the measured strain (in blue), as shown in Figure 1. These strain gauge histories look similar from this zoomed-out view or when viewed seperately. Statistically too, they look similar, with max and min strains around 375 and -170 microstrains. Let’s try zooming in to take a closer look say between 5400 to 6400 seconds and look at them seperately and also overlayed (Figure 2 and 3). You can see there are some differences in peak magnitudes, but for the most part, the waveforms track together to see how the strains are phased. This phasing is a very important attribute in correlating virtual and measured strains. If the phasing of the two is the same, then we know that FE results were modeled properly with the right boundary conditions, appropriate loads, polarities, and constraints. If the phasing is off, some key input that drives stress and strain has either been modeled incorrectly or left out. Further, if a cross plot option (Figure 4) is selected in the nCode tool menu, the time element is now removed and we can compare measured strain as a function of virtual strain. In an ideal plot, if a perfect linearlinear relationship existed, then this would a straight line but the lack of correlation appears as random scatter in data. This data shows some scatter with the randomness observed in Figure 4.
To summarize, if an FE model has been properly meshed, loaded, and constrained, the predicted virtual strains from nCode DesignLife at nominal stress regions would correlate with the measured strain histories. The degree of correlation can be quantified by comparing these actual-predicted values as illustrated in this example.
Best Practices for Virtual Strain Gauges
Finite Element Analysis (FEA)-based structural simulations, used to assess fatigue life, rely heavily on realistic modelling of the part. Many aspects in the modelling such as boundary conditions, material behaviour, dynamic characteristics, contacts, element type and size, etc. play a role in correctly assessing the local structural response. So far, the virtual strain gauge was the only way to correlate the calculated strains with the measured strains.
A new feature called Virtual Sensor has been introduced in nCode 2018. Virtual Sensors can be applied to the FE model to extract displacements at user defined locations. The virtual sensors can be applied interactively on the model. Once positioned, the Sensor normal is automatically aligned with the surface normal. The uniaxial or triaxial displacement time histories due to applied loads are then extracted and used for direct correlation with some real measured displacement data. Virtual Sensors provide complementary information to the Virtual Strain Gauges. The correlation of displacements enable a more global validation of the FE model, which is typically useful to validate the modelling of the boundary conditions and the mass and stiffness distributions.
For example, in dynamic conditions, it is imperative to understand if the relative motion between parts can come in contact especially when part nears or is at natural frequencies. In a vibration shaker table test rig, this could be reflected in accelerometer or displacement sensor data where large displacements at or around resonance conditions could be observed. When the component reaches resonance conditions, these sensor results along with input excitations can help understand the system level response under resonance conditions. Also in complex multi-component systems where different parts interact with each other, multiple dynamic events can be concurrently occurring over the test duration. In this case, it might also be difficult to position accelerometers (results integrated to get displacements) at various intricate locations to monitor real-time accelerations or displacements.
The predicted virtual displacement histories from Virtual Sensors can either be correlated with actual measured data or these predictions can be used to suggest design changes for parts. Figure 5 shows an example of predicting displacement histories at a node using the Virtual Sensor tool.
In this article, we looked at two powerful correlation tools available in DesignLife that highlight the ability to account for multiaxial load cases and how virtual strains or displacements can be recovered in user-defined location and direction for comparison with measured test results. The overall impact of utilizing correlation tools reduces the number of design and test iterations, increasing confidence of FE predictions, saving development time and cost.
Contact us to learn how you can implement nCode tools for correlating measured data with CAE results. Additional correlation tools include Strain Gage Positioning, Loads Reconstruction, and Modal Assurance Criterion (MAC Analysis).
Video - Correlating Experimental Test Data and FE Results
Video - FEA-Test Comparison with Modal Assurance Criterion (MAC) Analysis
Video - The Minimum Vital Signal Processing Skills an FE Simulation Engineer Must Have