Comparative analysis of artificial neural network model and analysis of variance for predicting defect formation in plastic injection moulding processes

Abstract

Abstract This study investigates the impact of plastic injection moulding process parameters on overflow defect formation. Experiments were conducted using a Taguchi L27 orthogonal array design. Multilayer Perceptron (MLP) artificial neural networks is explored and compared with ANOVA predictions. To assess model performance, the Root Mean Squared Error (RMSE) and the coefficient of determination (R2) is applied. The study considered temperature, speed, pressure, and packing force when constructing the MLP model using the back-propagation algorithm in Python. Results show that among the configured MLP neural networks, the 3-layer MLP architecture with sigmoid activation functions in hidden layers and a linear function in the output layer exhibited the lowest prediction error and the highest coefficient of determination. Comparative analysis reveals that the MLP neural network outperforms the ANOVA model, indicating superior prediction accuracy. The predicted outcomes from the ANN align well with experimental values, demonstrating the effectiveness of the ANN model in forecasting defect formation under specific process conditions. This research sheds light on the significance of process parameters and showcases the potential of MLP neural networks as a valuable tool in predicting and mitigating overflow defects in plastic injection moulding.