Material Model
Using practical test data to calibrate materials models for PET, POM, PP, and HDPE for use in finite element analysis.
A group project completed during 4th year of University.
Teammates: Fang Te Fong, Afdhal Zofran bin Mohd Amin, Nur Amanda binti Mustapha Kamal
Industrial Partner: Crux Product Design
Purpose
The primary purpose of creating the material models is to enable finite element analysis. For medical devices, and particularly for the project's industrial partner, streamlining the design process is essential. Reliability is also a high priority. This normally requires rigorous experimental testing of many iterations of multiple design concepts. However, with a verified, reliable material model, these expensive, time-consuming tests can be replaced by FEA.
Of course, FEA cannot fully replace physical testing, but the amount required can be significantly reduced.
Constitutive Models
Due to the inconsistencies in the data gathered by the different tests, not all of it could be combined to form one material model. Instead, for each material, two models were made.
TNV Model (Uniaxial and Drop Tower tests)
The first was calibrated using the uniaxial and drop tower impact test data and used a Three-Network Viscoplastic (TNV) constitutive model. Each network in the TNV model consists of a hyperelastic spring to capture elastic behaviour, and a viscoplastic flow response to capture dissipative viscous behaviour. This type of constitutive model is capable of capturing strain rate dependence, pressure dependence, volumetric plastic flow, and damage accumulation. This means it is still able to accurately predict behaviour when the strain on the material is high. However, as it doesn't access the DMA test data, it has a smaller sample size for predicting strain rate dependant material response and it has no data relating to the effects of temperature.
LVE Model (DMA)
To fill in for the weaknesses in the TNV model, another material model was created for each of the polymers. It was calibrated solely using the DMA data. A Linear Viscoelastic constitutive model was used, which comprises three parallel sets of hyperelastic springs in series with linear dampers. This type of model predicts strain rate-dependent behaviour very well, but doesn't take damage accumulation into account, limiting its effectiveness to only situations with low strains.
Example Calibration Results
TNV Model (Uniaxial and Drop Tower tests)
Comparison between possible constitutive models (POM)
The decision to use a TNV model with the drop tower and uniaxial test data came from this fitness analysis. Judging by both the maximum error between the actual and predicted results, and the Normalised Mean Average Difference (NMAD), the TNV mode performed best. This was true across all four materials.
Comparison between predicted and experimental data for PET
Comparison between predicted and experimental data for HDPE
Between all four materials, the TNV model fit the experimental results well. The main errors arose at large strains at moderate strain rates (500mm/min uniaxial test). This is likely due to the model not fully capturing the larger yield evolution exhibited at higher strain rates.
LVE Model (Dynamic Mechianical Analysis)
Comparison between predicted and experimental data for PET
Comparison between predicted and experimental data for HDPE
Across all four materials, the LVE model fit the experimental data well. The main errors arose from spikes in the experimental data that couldn't be smoothed out by MCalibration's smooth function. These spikes are almost certainly due to experimental error and the model's attempts to replicate them may cause issues when using the model for finite element analysis.