Wear Prediction of Polymers and Composites Using an Artificial Neural Network Approach
Final Report Abstract
Through methodology studies using pre-existing datasets of short fibers reinforced PA composites collected from the published literature and commercial databanks, it was demonstrated that an ANN model can be developed in order to predict accurately enough the mechanical properties and the tribological characteristics, i.e. friction coefficient and wear rate. The effects of training algorithm, architecture, number of outputs on the performance of the ANN network were evaluated and optimized in this project. It was revealed that a simple structure of one hidden layer including several neurons has an adequate capability to model straightforward relationships. However, complicated phenomena e.g. wear behavior, require more complex network topology. The determination of the best neuron number is case dependent. The prediction quality increases in a certain range when the number of neurons increases, yet it deteriorates when the neuron number exceeds a given saturation value. Usually, one-output network can avoid the disturbance between inputs and different outputs, and provide a higher prediction quality. Nevertheless, a multi-output network might be used for some simple problems, such as prediction of certain mechanical properties. The sliding friction and wear behavior of PPS based composites filled with various traditional fillers and inorganic sub-micro particles was both experimentally studied and modeled by implementing the ANN approach. A favorable synergetic action was found between the short carbon fiber (SCF) reinforcement and sub-micro TiO2 (300 nm) particles for reducing the friction and wear of PPS. When the particles are gradually released, they tend to mix with soft matrix material and Fe from the steel counterface giving rise to quasispherical transfer particles that serve as spacers and apparently roll between the two mating surfaces. The addition of either Gr or Gr in combination with PTFE effectively eliminated the stick-slip sliding motion of the hybrid system PPS/SCF/TiO2 in the high pv-range. The best results were obtained by blending comparatively low amounts of Gr and PTFE (≈ 5 vol.% for each). The solid lubricants acted in this case as lubricant reservoirs, similarly to the lubricants used in real ball bearings, and generated the lowest and almost constant frictional coefficient in steady state. The multiphase systems PPS/SCF/TiO2/Gr/PTFE show a great potential in real applications as sliding materials, especially for high pv conditions. The predicted surface plots by the trained ANN and the actual data exhibited very good conformity. These results illustrated the power of the neural network approach to predict the evolution of characteristic tribological parameters as a function of material type, applied pressure and sliding speed without having to perform the total number of experimental combinations. It should be noted that once the network was trained, the time required to output results for a given set of input was nearly instantaneous on a personal computer. The ANN methodology as developed will provide useful, less time- and cost-consuming tool for the design of high-performance tribo-composites.
Publications
- Prediction on Wear Properties of Polymer Composites with Artificial Neural Networks. Composites Science and Technology, 67 (2007) 168-176
Jiang, Z.; Zhang, Z.; Friedrich, K.
- Neural Network Based Prediction on Mechanical and Wear Properties of Short Fiber Reinforced Polyamide Composites. Materials and Design, 29 (2008) 628-637
Jiang, Z.; Gyurova, L.; Zhang, Z.; Friedrich, K.; Schlarb, A.K.
- State-of-the-art: On the Action of Various Reinforcing Fillers and Additives for Improving the Sliding Friction and Wear Performance of Polymer Composites. Part 1: Short Fibers, Internal Lubricants, Particulate Fillers. Journal of Plastic Technology, 4 (2008) 1-31
Gyurova, L.A.; Schlarb, A.K.
- Study on Friction and Wear Behavior of Polyphenylene Sulfide Composites Reinforced by Short Carbon Fibers and Sub-Micro TiO2 Particles. Composites Science and Technology, 68 (2008) 734-742
Jiang, Z.; Gyurova, L.A.; Schlarb, A.K.; Friedrich, K.; Zhang, Z.
- Modeling the Sliding Wear and Friction Properties of Polyphenylene Sulfide Composites Using Artificial Neural Networks. Wear, 268 (2010) 708-714
Gyurova, L.A.; Miniño-Justel, P.; Schlarb, A.K.
- Artificial Neural Networks for Predicting Sliding Friction and Wear Properties of Polyphenylene Sulfide Composites. Tribology International 44 (2011) 603-609
Gyurova, L.A.; Friedrich, K.
