ScholarWorks@경상국립대학교

초록

This study utilizes artificial neural networks (ANN) to model and predict surface roughness during minimal quantity lubrication (MQL) turning of Incoloy 800H. Real time turning experiments were conducted to measure the surface roughness under varying machining conditions. A backpropagation neural network (BPNN) model is utilized for the modeling process. The ANN architecture consists of one output neuron representing surface roughness and three input neurons corresponding to cutting speed, feed, and depth of cut. The experimental datasets were divided into three subsets: training, testing, and validation, in a 5:1:1 ratio. Statistical parameters, including mean squared error (MSE) and average error in prediction (AEP), were calculated to identify the optimal network configuration. The optimal network, with a 3-13-13-1 architecture, provides highly accurate surface roughness estimates, demonstrating excellent agreement between the ANN predictions and the experimental results.