Evaluation of machine learning methods in predicting optimum tensile strength of microwave post-cured composite tailored for weight-sensitive applications

Abstract

The study evaluated the performance of various machine learning methods in predicting tensile strength of microwave post-cured composite tailored for weight-sensitive applications. Using forty six training data pairs from the Box-Behnken design plan, an Adaptive Network-based Fuzzy Inference System (ANFIS) and Artificial Neural Networks (ANN) were built to predict the optimum tensile strength of polyurethane wood ash composite. Numerical optimization was done using the graft of ANFIS-MOGA method. The process control factors considered were particle size, curing time, power level, volume fraction and curing angle. The predictive accuracy of the evaluated machine learning methods were assessed using Coefficient of Determination (R2), Root Mean Square Error (RMSE), Mean Average Error (MAE) and Standard Error of Prediction (SEP). RSM (RMSE = 0.0339, MAE = 0.0002, SEP = 0.0295, R2 = 0.994) and ANFIS (RMSE = 0.0307, MAE = 0.0098, SEP = 0.0267, R2 = 0.995) models gave higher degree of accuracy than ANN model (RMSE = 0.0827, MAE = 0.0124, SEP = 0.0719, R2 = 0.988). The optimization exercise gave optimal tensile strength of 2.27 MPa at optimum process setting of 177µm particle size, 33 min of curing time, power level at 411 Watt, 42% volume fraction and 11° curing angle. Complementary trial at the named optimum process setting conveyed workable results. Furthermore, selecting microwave post-cured composite for weight-sensitive applications was justified considering that the desirability factor for polyurethane (42% wood ash) is fairly higher than other material in the same class, this points to the fact that deployment of the microwave post-cured composite in weigh-sensitive applications could benefit weight reduction.