In-situ monitoring and regulation of surface shape in swing fixed abrasive lapping of BK7 glass

Abstract

Fixed abrasive lapping is a critical process that affects the surface accuracy of optical components, its machining accuracy directly determines the overall imaging performance of optical parts. However, the surface shape of the workpiece during lapping is time-varying, and the convergence timing is judged by the operator's experience and offline measurement results, which introduces great uncertainty into subsequent process control. To acquire the surface shape status of the workpiece and improve its controllability, this study established a multi-source information acquisition platform for swing fixed abrasive lapping (SFAL). The spindle motor current signal U1, eccentric wheel motor current signal U2, and acoustic emission (AE) signal were fused at the feature level based on multi-sensor information fusion technology. A random forest classification model optimized by the sparrow search algorithm (SSA-RF) was employed to achieve in-situ monitoring of workpiece surface shape category (convex, flat, concave). Based on the kinematic model of SFAL, the distribution of abrasive sliding distance under different parameter combinations was investigated, and process regulation strategies were proposed for convex and concave workpieces. The results indicated that when the fused signal features were used as input, the prediction accuracy of the SSA-RF model improved by more than 15 %, and a classification accuracy of 89.83 % for workpiece surface shape was achieved. For workpieces that do not meet the convergence condition, process regulation can be adopted to change the distribution of abrasive sliding distance on the workpiece surface, thereby facilitating the evolution toward surface flattening. After process regulation, the peak-to-valley (PV) value of the convex workpiece surface profile converged to 1.72 μm; The surface shape convergence efficiency of the concave workpiece in the early processing stage was improved, and the surface profile PV value finally converged to 1.29 μm. This study provides a theoretical foundation and technical approach for the high-precision machining and intelligent development of optical components.