Hybridized Artificial Neural Network-Based Expert Systems for Modelling of Robotic- Wire and Arc Additive Manufacturing Process

For proper condition monitoring and automation of the robotic- wire and arc additive manufacturing process, establishment of accurate relationship between the inputs and the responses is extremely important in both forward and backward directions. In this study, both feed-forward neural networks and radial basis function networks have been developed and compared for the aforementioned purposes. These models have been hybridized with some ancient metaheuristic optimization algorithms such as genetic algorithm and particle swarm optimization, as well as some rare and newly evolved ones such as firefly algorithm, covariance matrix adaptation evolution strategy, cuckoo search, bio-geography-based optimizer and grey wolf optimizer for the training purpose. Also, the performances of various algorithms have been compared with the help of some statistical methods. The novelty of the present study lies in hybridizing the feed-forward network with the said rare and newly evolved optimization algorithms and in utilization of radial basis function network for developing predictive models in the domain of robotic- wire and arc additive manufacturing process. The grey wolf optimizer tuned neural network has been found to produce most accurate models with feed-forward architecture while the radial basis function network trained with pseudo-inverse has outperformed all the other approaches considered in the study.