Performance Sensitivity Analysis of Natural Mineral Silicates Filled Friction Composites Using GD-ANN Approach

Abstract

Abstract Structurally dissimilar clay type silicate (tubular halloysite, platy montmorillonite, and acicular wollastonite) reinforced compression molded composites were fabricated with and without MgO. The thermo-mechanical and frictional performances of natural silicate-based friction materials were systematically evaluated with respect to the silicate and silicate-MgO free composites. The morphology and hardness of the natural silicates dominated the mechanical and tribological responses of the friction materials. The worn surface morphology revealed the influence of natural silicate on braking dynamics as evident from performance sensitivity analysis, braking load, sliding speed, and temperature. The nature of sliding induced tribo-layers was ascertained from elemental mapping by EDX and Raman spectroscopy indicating the friction composition of tribo-layer to be influencing performance sensitivity. Under the variable operating conditions, halloysite-based friction materials showed excellent wear resistance, and wollastonite-based friction composites with MgO enhanced the friction coefficient (∼0.43–0.61) while exhibiting minimum load-speed sensitivity. The gradient descent learning algorithm-based artificial neural network (GD-ANN) with optimally tuned network architecture predicted (R2 ∼ 97%) both the tribological performance attributes (coefficient of friction and specific wear-rate) of the natural silicate-filled friction composites more accurately as compared to the conventional regression analysis.