Recurrent neural network based analysis for laser cladding dynamic model identification

In this article, a recurrent neural network is applied to dynamic modeling of the laser cladding by powder injection. The recurrent neural network takes into account table velocity, laser pulse energy, laser pulse frequency, and laser pulse width as the input parameters, and the clad height and rate of solidification as the outputs. A large set of data obtained from experimental studies with different settings is fed to the network for training. The backpropagation through time method is used to train the network. Unseen data in the training process are used for verification. Results are promising and show that the identified model is able to predict the nature of the laser cladding process in both transient and steady state responses.