In order to solve the problems of low efficiency and poor precision caused by the traditional die-cast structural components allowance detection relying on manual positioning reference, an on-machine measurement method based on no-positioning reference was proposed. Through the integration of high-precision contact probe and five-axis CNC system, an intelligent machining platform with online measurement function was constructed to realize the closed-loop control of machining and measurement. The research focuses on the development of a dynamic trajectory planning algorithm for probes, an autonomous optimization of measurement paths based on topological correlation mapping of surface features, an adaptive measurement path planning strategy and an allowance compensation mechanism based on point cloud registration to realize the dynamic prediction and intelligent correction of machining allowances. The results show that the proposed feature-driven virtual reference construction method replaces the physical tooling with the intelligent feature matching, the established error propagation suppression model under process disturbance effectively improves the measurement stability under complex working conditions. The research results provide a highly flexible and high-precision allowance control solution scheme for intelligent manufacturing, which has significant engineering application value.