Artificial Neural Network Approach to Predict the Elastic Modulus from Dynamic Mechanical Analysis Results

Xianbo Xu, Nikhil Gupta

Research output: Contribution to journalArticlepeer-review


Characterizing viscoelastic materials over a range of temperatures and strain rates requires an elaborate experimental scheme. Methods are available that allow testing a single specimen and then transform the storage modulus data to recover elastic modulus. However, applying these methods to polymers with more than one thermal transitions is challenging. As the form of artificial neural network (ANN) follows the time–temperature superposition principle, it is used in the present work to establish the master relation for storage modulus. The neural network is built with various neuron numbers and regularization factors, and two magnitudes of Gaussian noises. The predictions achieve the best accuracy when the regularization factor equals 10−4 and neuron number equals the number of thermal transitions in the material. Then the ANN is trained on storage modulus data of graphene-epoxy nanocomposites and achieves 4.1% average error. The storage modulus is transformed to time domain relaxation function using integral relation of viscoelasticity. Stress response over the strain history is determined and the elastic modulus is extracted. Compared to the tensile test results, the predictions achieve an average error of 0.7%, which indicates that the method can predict the material behavior over a wide range of temperatures and strain rates.

Original languageEnglish (US)
Article number1800131
JournalAdvanced Theory and Simulations
Issue number4
StatePublished - Apr 1 2019


  • dynamic mechanical analysis
  • neural network
  • ridge regression
  • viscoelasticity

ASJC Scopus subject areas

  • Statistics and Probability
  • Numerical Analysis
  • Modeling and Simulation
  • General


Dive into the research topics of 'Artificial Neural Network Approach to Predict the Elastic Modulus from Dynamic Mechanical Analysis Results'. Together they form a unique fingerprint.

Cite this