A robot-mounted machine vision system for sub-millimeter accurate localization of scribe line and trim-edges on composite parts

This paper presents a robot-mounted machine vision solution to achieve sub-millimeter localization accuracy of scribe lines and trim-edges on composite parts. In current practice, post-cure trimming to achieve a desired part shape is typically performed on large CNC gantry mills with complex fixtures, which are costly and inflexible. Six DOF industrial robots offer a more flexible platform, but part-to-part geometric variation and limited robot accuracy make it difficult to guarantee a desired trim tolerance in the sub-millimeter range. Scribe lines embedded in layup mandrels are used in practice to provide visual indication of the desired final part edge shape. Therefore, accurate localization of this scribe line feature is essential to perform robotic trimming of the composite parts along the scribe line. We present a telecentric camera-based imaging system mounted on a 6 DOF robot and an image processing pipeline to localize and measure the distance between the scribe line and part edge with sub-millimeter accuracy of less than ±0.5 mm. We also introduce a novel feature detection method which utilizes a finetuned monocular depth estimation deep neural network to generate relative depth images and enhance part-edge contour detection for challenging textures, surface anomalies, and varying edge and scribe line quality. The system is evaluated on carbon fiber reinforced polymer (CFRP) composite samples under dynamic scanning at 10, 25, and 50 mm/s speed. The proposed solution achieved an overall mean measurement error of −0.068 mm with a standard deviation of 0.038 mm in estimating the scribe line to trim-edge distance. The results demonstrate that the proposed robot-mounted machine vision system and algorithm can meet target trim tolerance requirements and enable perception informed robotic machining of large composite structures.