Multi-Input Multi-Output Modeling of Anthill Clay-Bonded Sand Mold System Using Artificial Neural Networks: Forward and Reverse Predictions

Abstract

Anthill clay sand molding is characterized as a multi-input, multi-output system where molding sand properties, such as permeability, compression strength, collapsibility, and mold hardness, are affected by an array of process parameters, including anthill-to-sand ratio, number of strokes, mulling time, water content, and coal dust. This interdependence of various parameters is often poorly captured using conventional experimental approaches and statistical modeling, limiting process control accuracy. ANN-based modeling is used to predict and optimize sand mold properties. This study compares two ANN models: the Back-propagation Neural Network (BPNN) and the Genetic Algorithm Neural Network (GA-NN) for forward and reverse modeling of the anthill clay bonded sand mold system. Forward modeling predicts sand mold properties by known process parameters, while reverse modeling objectively determines optimum process conditions for the desired properties of the mold. The models are trained using 1000 experimental datasets and analyzed based on statistical performance metrics such as mean absolute percentage error (MAPE), root mean square error (RMSE), and correlation coefficient (R). Results indicate that the GA-NN model is superior to BPNN, showing prediction accuracy and less percentage error for all mold properties in forward predictions and molding sand variables in reverse predictions. The present GA-NN model is a remedy for multi-collinearity issues in mold property prediction, rendering the model more reliable for prediction. It can also be considered as a platform for real-time monitoring and optimizing sand molding processes. This study provides insights into how AI-driven modeling techniques can enhance the quality of castings and the efficiency of processes in the foundry sector.