Effect of machining parameters on surface roughness for compacted graphite cast iron by analyzing covariance function of Gaussian process regression. (June 2020)

• Record Type: Journal Article
• Authors: Lu, Juan; Zhang, Zhenkun; Yuan, Xuepeng; Ma, Junyan; Hu, Shanshan; Xue, Bin; Liao, Xiaoping
• Is Part Of: Measurement. Volume 157(2020)
• Publication Date: 2020-06
• Subjects: Compacted graphite cast iron -- Surface roughness prediction -- Gaussian process regression -- Correlation analysis; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2020.107578
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Prediction model of surface roughness for Compacted Graphite Cast Iron is developed by Gaussian process regression (GPR). The effect of training set size on prediction performance of GPR model has been evaluated. Prediction performances of GPR model with and without k -fold cross-validation are evaluated. The effect of machining parameters on surface roughness is analyzed from the covariance function of GPR. Abstract: This study employs Gaussian process regression (GPR) with square exponential covariance function to predict surface roughness of Compacted Graphite Cast Iron (CGI). In addition, a comparative study is conducted on prediction performances for GPR with and without cross-validation, back propagation neural network (BPNN) and support vector machine (SVM) for milling experiment of CGI. Experimental results indicate that prediction performances of GPR without cross-validation and GPR with cross-validation (GPRCV) are similar, and both superior to BPNN and SVM. The effect of machining parameters on surface roughness characterized via length-scale hyperparameters of covariance function is excavated according to prediction principle of GPR. The result shows that cutting speed and feed speed significantly affect surface roughness, depth of cut produces little impact on surface roughness within the given parameter intervals. The analysis of distance between test and training sets and 3D merged surface of surface roughness have verified that the effect is reasonable.