Application of Multiple Back-Propagation ANN for Predicting Finish Surface Roughness Produced by Vereco CNC Cylindrical Grinding Machine

Abstract

Large industrial operations, such as those in manufacturing, marine, and oil & gas, rely on cylindrical grinding as one of their most critical machining processes. One of the difficulties encountered throughout the cylindrical grinding process is anticipating the workpiece’s exact surface roughness after the process. Due to the process’s inability to predict the exact surface roughness, the processing time and quality produced are difficult to control. Several independent variables that are immediately quantifiable are used to build a data set for the training procedure in this study. Predicting the ultimate surface roughness generated by the cylindrical grinding process is critical for optimizing production time, quality, efficiency, and customer satisfaction. An artificial neural network with multiple backpropagation algorithms is applied. Through the learning process, the best combination of learning is obtained, namely: a learning rate of 0.057 and a momentum of 0.434 with one hidden layer in which there are 10 hidden nodes. This combination is believed to be the best training combination to produce the minimum error between the target and the true value. The root mean squared error of the test calculation was 0.0436 with a prediction accuracy of 95.64%. This set of experiment results produces predictive results through the validation process and succeeds in predicting the finish surface roughness with promising results (accuracy in the range of 94.683-97.661%).