Artificial neural network modeling to evaluate polyvinylchloride composites’ properties

  • aIndustrial Engineering Department, German Jordanian University, Amman, Jordan
  • bMechanical Engineering Department, German Jordanian University, Amman, Jordan
Received 22 January 2018, Revised 1 June 2018, Accepted 2 June 2018, Available online 15 June 2018


ANN modeling for the nonlinear constituents-properties relationship of PVC composites.

Identification of pertinent training algorithm and activation function.

Optimal PVC composites’ constituents’ percentages settings for pre-required properties.


The mechanical properties of extruded Polyvinylchloride (PVC) composites cannot be easily predicted due to the nonlinear nature of the relationship between the composite’s composition and the resulting after-production properties. In the work presented herein, supervised artificial neural network (ANN) modeling is used to predict and optimize three properties (tensile strength, ductility and density) of PVC composites having different weight percentages of virgin PVC, CaCO3, plasticizers, and recycled PVC. Different ANN models, designed and analyzed through factorial design and analysis of variance methodology, were evaluated using an experimental dataset which was designed according to the mixture design of experiments approach. The results show that the constituents-mechanical properties relationship of PVC composites’ can be accurately estimated using several ANN models including Levenberg-Marquardt/6-[18-9]3-3/Radial basis and Levenberg-Marquardt/6-[9-18]3-3/Log sigmoid models. The results also show that the ANN modeling is capable of determining the optimal weight percentages of the different PVC composite constituents in order to achieve a required composite property.


  • Polymers and plastics;
  • Artificial neural networks;
  • Mechanical properties;
  • Supervised learning;
  • Design of experiments