摘要
The thin-walled lens barrel is a crucial component in high-precision optical systems and must be machined with high dimensional accuracy and surface quality. However, in the machining of such lenses, variations in material removal rates and surface blemishes can stem from machining parameter settings, tool wear, chip tangling, and tool offset error, resulting in difficulties in controlling workpiece dimensions and surface quality. Consequently, manual inspections and troubleshooting are required, leading to labor-intensive procedures and unscheduled production disruptions that compromise efficiency. To address these challenges, this study developed an expert system based on multi-sensor data fusion for real-time thin-wall machining. Though the use of acoustic emission and three-axis accelerometer sensors installed on a turn-milling machine, pivotal machining parameters such as cutting speed, feed rate, and depth of cut were gathered and analyzed. Support vector regression and artificial neural network models were developed for monitoring material removal rates and tool and chip status, respectively. Principal component analysis was used to increase data efficiency. The outcomes were promising, with the model achieving 95.39 % accuracy for predicting the material removal rate under normal processing conditions and 89.63 % accuracy under abnormal conditions. The root-mean-square error (RMSE) of the model was 0.2568cm 3 /min. By extrapolating product dimensions using a specific formula, the achieved accuracy reached 0.009 mm. The tool status and chip status predictive models exhibited RMSE values of 0.002904 and 0.001039, respectively. This integrated approach can be used to effectively diagnose and rectify defects during the production of optical lenses. It allows for swift dimensional detection, enabling operators to implement accurate solutions, thereby enhancing yield and efficiency.