Combining Experiments, PolyUMod, Kornucopia, and Abaqus to Create Accurate FE Scratch Simulations
DuPont
Ted Diehl, Li Lin, Ye Zhu, John J. Podhiny, Richard T. Chou, and Jeffrey A. Chambers
Veryst Engineering, LLC
Jorgen S. Bergstrom, Xiaohu Liu 2013 SIMULIA Community Conference
Overview
Goal of Project Complexity of Scratch Material Characterization Material Testing Constitutive Approach, and Calibration FEA Modeling Approach To Simulate Scratch Correlating 3D Physical Scratch Data with FEA Simulation Results Open Issues More To Do! Conclusions
Project Goals
Develop Abaqus models to predict scratch performance of polymeric materials Predict residual depth and residual shape of scratch. Including time dependent partial self-healing behavior of many polymers. Material Characterization Include nonlinear viscoplastic nature of polymers via PolyUMod. Utilize efficient Ziggurat testing protocols and MCalibration for material calibration. Develop improved methods to compare 3D physical scratch test data to Abaqus scratch simulation results. Utilize Kornucopia analysis tools to process and transform large amounts of messy, 3-D data for correlations.
Polymer Materials Used in Study
Three polymeric materials were studied. DuPont Surlyn grades (an ethylene copolymer ionomer). Surlyn 9950 Surlyn 1706/1707 blend Materials are differentiated from each other by such factors as their percent neutralization, acid content and other formulation make-up. PMMA A polymethylmethacrylate. PMMA is noticeably different from the Surlyn materials. PMMA is stiffer and harder. PMMA is based on the molecule methyl methacrylate and is not an ethylene copolymer ionomer.
Complexities of Polymer Scratch
Photographs of 3 different polymers after being scratched by the same indenter
Indenter load progressively increased
Complexities of Simulating Polymer Scratch
What about Polymer fracture? The following are required for viable simulations and scratch validation. This is important, but was not Material experiments that exercise the sample to generate sufficient data to considered in this initial study. characterize elastic, inelastic, and recovery behavior, including rate/time dependence and large strain response. A polymer constitutive model capable of representing such complex behavior. Robust FEA methods that can incorporate the advanced material model, support sliding contact, and can handle extremely large strains which will cause severe element distortions. Detailed polymer scratch experiments with 3-D surface imaging technology to capture the scratched profile for comparison to the FEA simulations. A method to manipulate and correlate large amounts of data from tests and models. Including a method to correlate 3-D scratch surfaces.
Material Characterization
Physical Tests and Constitutive Models
The Ziggurat Test Protocol
An efficient all-in-one material testing protocol with these key benefits: Exercises elastic, inelastic, loading and unloading behavior. Adds stress relaxation behavior at several strain levels in same test Can be applied to compression (shown), tension, shear, equibiaxial Applied Boundary Conditions
Compressive Strain
Materials Response
Time
Compressive Stress
Compressive Strain
Application of Ziggurat Test Protocol to Polymers with Significant Plasticity
The previous slide presented an idealized Ziggurat protocol. Elastomer testing often looks quite similar to picture on right. Tests of polymers that exhibit noticeable inelastic (plasticity) behavior often need a modified ziggurat protocol. Key goal of Ziggurat protocol is to get several relaxation holds on both the loading and unloading curves. Depending on relaxation behavior of material, time holds might be very long compared to loading ramps of various strain segments.
10
Comp. engineering stress (MPa)
Ziggurat Protocol (Surlyn 1706/1707)
Key items Asymmetric strain/time BC Very long time holds Difficulty with strain control (cross-head) vs. laser gage strain measurements. Laser measures used for material calibration. Idealized Protocol
Comp. engineering stress (MPa)
150
150
100
100
50
50
0 0 10 20 30 40 50 60 Comp. engineering strain (%) 70
0 0 40 50 60 10 20 30 Comp engineering strain (%) 70
a) Strain computed from cross-head displacement
Comp eng. strain (%) 60 50 40 30 20 10 0 Comp. eng. stress (MPa) 150 125 100 75 50 25 0 0 10 0
b) Strain computed from laser gages measuring specimen height
Strain computed from laser gages measuring specimen height 10 20 30
Time (minute)
20
30
c) Strain and stress data as a function of time
11
Ziggurat Protocol (Surlyn 1706/1707)
Key items (continued) Raw data (not shown): 35,000 points, non-uniform t. Strain data had a small amount of noise due to optical laser gages. Data had start-up distortions in the initial small-strain regime due to small geometry irregularities in the samples. Kornucopia utilized to clean-up and re-sample data prior to MCalibration analysis. This improves both accuracy and efficiency of material law calibration process.
12
Ziggurat Protocol (Surlyn 1706/1707)
Raw data clean-up steps: Semi-automatic algorithm separates data into various Ziggurat segments. Strain holds and all the other segments Each segment smoothed to remove noise Distortion at the beginning of the test was trimmed and corrected with linear extrapolation. Each segment resampled Load/unload segments 100 equal time points Strain holds 100 logarithmically spaced pts. Data clean-up significantly improved the data quality and reduced the dataset from 35,000 points to less than 1,000 points. Reusable worksheet allowed for efficient, repeatable, and traceable data cleanup of all Ziggurat tests for all materials.
13
PolyUMod Three Network Model (TNM)
Material data exhibits non-linear viscoplastic response with significant flow and partial recovery after unloading. Utilize PolyUMod library of user-material models for Abaqus The Three Network Model is a rheological approach with non-linear springs and dashpots capable of representing materials. See papers references for details of the governing equations PolyUMod and MCalibration are software products from Veryst Engineering, LLC.
14
PolyUMod Three Network Model (TNM)
Total stress is given by sum of the Cauchy stress tensor in each network. Stress response of each of the 3 networks Hyperelastic Arruda-Boyce eight-chain model. The stiffness of network B is taken to evolve with the plastic strain accumulation in Network A. The rate of viscoplastic flow of Networks A and B is given by a power law expression of the driving deviatoric shear stress and temperature, and the flow resistance is taken to be pressure dependent. This material model exhibits true plasticity despite the presence of the Network C component. The cause of the plasticity is the energy barrier for flow that is inherent in the flow equations.
15
Calibrating TNM with Test Data via MCalibration
Utilize Compression with Friction load case, MCalibration automatically creates an Abaqus FE model of the experimental test setup and boundary conditions (including friction). Allows model to closely represents the configuration of actual tests.
16
Calibrating TNM with Test Data via MCalibration
MCalibration performs nonlinear search to find material law parameters that minimize error between model and actual test data. Several controls and options to allow user influence over process.
TNM Representations of Material Test Data
17
TNM Representations of Material Test Data
18
19
Abaqus Simulation of Scratch
20
Abaqus/Explicit Model to Simulate Scratch
Key features & approach Explicit model selected to more robustly handle potential severe deformations. 100 micron radius scratch tip rigid surface Polymer specimen was divided into two regions (tied together) Coarse region (500 C3D8R), Fine region (55,000 C3D8R). Adaptive re-meshing (by solver) used for fine mesh region. Mass scaling is used to speed-up the simulation. More to be said about this shortly.
21
Abaqus/Explicit Model to Simulate Scratch
The scratch simulations were performed in three steps: (example for Surlyn 1706/1707 blend simulation) 1. Indenter pushed vertically down into material 60 m in 0.05 sec. 2. Indenter moved horizontally 0.6 mm in 0.15 sec. 3. Indenter unloaded by moving vertically upward 66 m in 0.05 sec. Allow material to creep recover (>20 seconds in the FEA model)
Step 3
MP4 GIF
Step 1
Step 2
Notes:
Different scratch depths used for other materials Simulations were performed in displacement control
Physical scratch experiments were performed in load control
22
Abaqus/Explicit Model to Simulate Scratch
Displacement (mm)
Mass Scaling Issues with Explicit Model
Compared to Courant stability of model, scratch time is long, especially for creep recovery. Use staged mass scaling to speed-up simulation. Caused large oscillations Minimized by massproportional Rayleigh damping. Segmented filtering by Kornucopia shows noticeable discontinuity caused by staged mass-scaling. Recent analysis shows we were just too aggressive!
Creep recovery of single node in scratch region
0 0.02 0.04 0.06
23
10
15
20
25
Displacement (mm)
0 0.02 0.04 0.06
Stage 1, t=0.029msec
Stage 2, t=0.3msec
0 1 2 3
Stage 3 t=3.0msec
4
Displacement (mm)
0 0.02 0.04 0.06
All stages, No smoothing Stage 1, smoothed Stage 2, smoothed Stage 3, smoothed
0 1 2 3 4
Displacement (mm)
0 0.02 0.04 0.06
Noticeable Discontinuity
0 5 10 Time (sec) 15
Stage 1, smoothed Stage 2, smoothed Stage 3, smoothed
20 25
24
Physical Scratch Tests and Correlations to Abaqus Simulations
25
How to Compare Experiments and FEA???
Visual image
Abaqus top surface
Zygo optical scan
26
Typical Physical Scratch Tests
Zygo 3-D optical surface profile (Surlyn 9950, 1.5N Load, ~ 1 day post scratch)
Many drop-outs in data Limited 3-D viewing Limited data analysis
27
Typical Physical Scratch Tests
Zygo 3-D optical surface profile (Surlyn 9950, 1.5N Load, ~ 1 day post scratch)
60
Improvements after Data Healing via Kornucopia and Mathcad
Z (micron)
50
40 Healed Raw 30 0 100 200 300 400 500
Distance (micron)
Further Scratch Test Data Analysis
(PMMA ~ 1 day post scratch)
28
Collapsing many surface cross-sections into a single average cross-section
Further Scratch Test Data Analysis
(Surlyn 1706/1707 blend ~ 1 day post scratch)
29
Collapsing many surface cross-sections into a single average cross-section
Similar Post-Process of FEA Scratch Results
(Surlyn 1706/1707 blend ~ 25 seconds post scratch) FEA postprocessing required additional surface remapping of results to enable slicing
30
Collapsing many surface cross-sections into a single average cross-section
31
Final Cross-Section Correlations
Average scratch cross-sections show that Abaqus/Explicit models using PolyUMod TNM material model correlates well across several diverse materials.
32
Open Issues
Polymer fracture Difficult to model as many fracture locations can occur. Need to assess viable routes for this very complicated. Creep recovery over long time periods The amount of mass scaling we used was shown to be a bit aggressive. Lesser mass scaling would be better, but makes even longer run times. Switching from Abaqus/Explict back to Abaqus/Standard not possible because of adaptive meshing approach used. We only computed for 25 seconds of creep recovery physical tests were for 1 day! A DSP concept Frequency Warping. Nonlinearly warp time-domain implication of visco material constants so as to compress, in time, the short and long scale meanings of constants. Similar to filter warping or structural frequency warping.
33
Conclusions
Demonstrated that accurate models of Polymeric Scratch are possible. Combined: Detailed physical material tests (Ziggurat Protocol) PolyUMod Three Network (material) Model for Abaqus MCalibration material calibration Abaqus/Explicit using mass scaling and adaptive re-meshing Kornucopia data analysis Methodology developed and demonstrated using three different polymeric materials, two DuPont Surlyn grades and PMMA. Demonstrated efficient method to compare complex 3-D scratch results Open issues Polymer fracture Improved methods to simulate long creep recovery times Something to explore Frequency Warping
34
