Efficiency assessment of ANFIS in surface topology prediction in diamond turning of RSA443 optical aluminum using Small Datasets

Abstract

This paper presents an assessment of the efficiency of the ANFIS model in surface topology prediction of RSA 443 optical aluminium using small datasets. The experiments are designed and conducted based on the Taguchi L9 orthogonal array. The single point diamond turning procedure has been carried out on Nanoform 250 Precision CNC machine. Cutting speed, feed rate and depth of cut together with acoustic emission signal root mean square, dominant frequency and peak rate are the ANFIS model input parameters. The function that gave the best results was the sigmoidal membership function, hence it is used in this paper. To evaluate model performance, the measured surface roughness results are compared to ANFIS predicted results based on the Mean Absolute Percentage Error (MAPE) and prediction accuracy. The MAPE values show that the ANFIS prediction accuracy of surface topology is 79.42%. This value is within the upper quartile range, indicating the high predicting accuracy of ANFIS even in the absence of enough training data. The model can be reliably utilized in surface roughness prediction using small datasets while its prediction accuracy can however be improved by using larger datasets.