Prediction of collapsibility of moulds and cores of CO2 sands using a neural network
The CO2 process of making sand moulds and cores is a well-established process and suitable for all types of foundry. However, the collapsibility of CO2 sand is quite poor. A variety of additives are used to improve collapsibility of CO2 sands. Several other process parameters also affect collapsibility of CO2 sands. In the present investigation an attempt has been made to use an artificial neural network (ANN) model for prediction of the collapsibility of CO2 sand. Experiments were conducted with various input process parameters, such as binder content, gassing time, temperature and additive content using three different additives, namely coal dust, dextrin and alumina. The objective of the experiments was to generate basic data to train a back-propagation ANN model and finally predict collapsibility in terms of retained compressive strength of CO2 sands for the test data. A three-layer neural network model with six input neurons corresponding to six input process parameters, one output neuron corresponding to collapsibility and 19 hidden neurons has been suggested, which gives a maximum error of 2 per cent in prediction of test data. Results indicate that prediction of the collapsibility of CO2 sand with an ANN model is feasible. Predicted values match experimental values quite closely.