In this paper, genetically evolved Artificial Neural Networks (ANN) built with sparse data for predicting the depth of penetration of Abrasive Water Jets (AWJs) into the material is proposed. Sparse data was collected during the cutting trials on Steel 1.4301 and AlMgSi0.5 alloy with AWJs considering various process parameters like jet pressure, abrasive mass flow rate, jet traverse rate, diameter of focusing nozzle, stand of distance, number of passes and type of abrasive material. The data was generated by employing abrasive water injection jet (AWIJ) and abrasive water suspension jet (AWSJ) systems. In developing ANN using conventional Back Propagation (BP) learning algorithm, random selection of parameters such as weights, learning rate parameter, momentum parameter is quite tedious and error prone. Hence, the proposed method attempts to select the weights by Genetic Algorithms (GA) in order to develop ANN in an optimal manner. Performance of the proposed method is compared with that of ANN built with BP learning algorithms and regression models, both built with abundant data. Finally, the effectiveness of the proposed method for situations with sparse data is demonstrated.
展开▼
