Classification of drilling surface roughness on computer numerical control (CNC) machine tools based on Mobilenet_v3_small_improved

Abstract

Computer numerical control (CNC) machine tool drilling is a crucial process in the contemporary manufacturing sector, facilitating high-precision fabrication of complex components and thus enhancing production efficiency and product quality. Surface roughness serves as a principal quality metric in machining operations. Spindle speed and feed rate are primary determinants influencing the surface roughness during the CNC drilling process. This study introduces data acquisition software developed on the Syntec CNC system and MySQL platform to enable real-time data capture and storage, setting a foundational dataset for subsequent analysis of roughness classification. Additionally, an enhanced roughness classification model using the improved MobileNet_v3_small model is presented. The model integrates dual time–frequency plot features of short-time Fourier transform (STFT) and continuous wavelet transform (CWT) to provide novel input features for the MobileNet_v3_small architecture, the output of which is a workpiece surface roughness classification. Fusing the time–frequency features of STFT and CWT serves to refine the classification capability of the network structure. Validation of the network model followed during training, giving training, validation, and test accuracies of 85.2 %, 84 %, and 85.4 %, respectively. Comparative analysis with other lightweight industrial network models reveals that the improved MobileNet_v3_small model demonstrates average accuracy enhancements of approximately 10 %, 9 %, and 13 % across the training, validation, and test datasets, respectively. Reductions in the root mean square error averaged 0.15. Experimental results indicate the superior classification accuracy of the improved MobileNet_v3_small model in drilling surface roughness.