A Hybrid Model for Predicting Sand Casting Defects

Casting defects cause loss of time for reworked items as well as loss of scrapped material. In sand casting, there are many parameters with interrelated complex interactions which have effects on the quality of the product. Artificial Neural Networks (ANN) are widely used for modeling the processes in the casting industry; but possible detection of the significant parameters before the application of an ANN should bring an advantage to obtain more accurate results—possibly in a shorter duration of time. Decision Trees (DT) can also be used to reduce the whole data set with respect to significant parameters. It is expected that the reduced data will yield a more efficient use of ANN to optimize casting parameters. Simulated data has been generated to compare the performance of the pure application of ANN versus a hybrid approach of DT and ANN. It can be observed that the hybrid approach gives a slightly better result than the pure ANN approach in terms of coefficient of determination (r-squared). The number of iterations required for the hybrid approach is much less than that of the pure ANN approach owing to the efficiency due to incorporation of the decision trees.