Abstract
A vision system is described which performs real-time inspection of dry carbon-fibre preforms during lay-up, the first stage in resin transfer moulding (RTM). The position of ply edges on the preform is determined in a two-stage process. Firstly, an optimized texture analysis method is used to estimate the approximate ply edge position. Secondly, boundary refinement is carried out using the texture estimate as a guiding template. Each potential edge point is evaluated using a merit function of edge magnitude, orientation and distance from the texture boundary estimate. The parameters of the merit function must be obtained by training on sample images. Once trained, the system has been shown to be accurate to better than ±1 pixel when used in conjunction with boundary models. Processing time is less than 1 s per image using commercially available convolution hardware. The system has been demonstrated in a prototype automated lay-up cell and used in a large number of manufacturing trials.