Multi-objective evolutionary optimization of polynomial neural networks for fatigue life modelling and prediction of unidirectional carbon-fibre-reinforced plastics composites

Abstract

In this article, evolutionary algorithms (EAs) are employed for multi-objective Pareto optimum design of group method data handling (GMDH)-type neural networks that have been used for fatigue life modelling and prediction of unidirectional (UD) carbon-fibre-reinforced plastics (CFRP) composites using input—output experimental data. The input parameters used for such modelling are stress ratio, cyclic strain energy, fibre orientation angle, maximum stress, and failure stress level in one cycle. In this way, EAs with a new encoding scheme are first presented for evolutionary design of the generalized GMDH-type neural networks, in which the connectivity configurations in such networks are not limited to adjacent layers. Second, multi-objective EAs with a new diversity preserving mechanism are used for Pareto optimization of such GMDH-type neural networks. The important conflicting objectives of GMDH-type neural networks that are considered in this work are training error (TE), prediction error (PE), and number of neurons ( N). Different pairs of these objective functions are selected for two-objective optimization processes. Therefore, optimal Pareto fronts of such models are obtained in each case, which exhibit the trade-offs between the corresponding pair of conflicting objectives and, thus, provide different non-dominated optimal choices of GMDH-type neural network model for fatigue life of UD CFRP composites. Moreover, all the three objectives are considered in a three-objective optimization process, which consequently leads to some more non-dominated choices of GMDH-type models representing the trade-offs among the TE, PE, and N (complexity of network), simultaneously. The comparison graphs of these Pareto fronts also show that the three-objective results include those of the two-objective results and, thus, provide more optimal choices for the multi-objective design of GMDH-type neural networks.