Intelligent detection of edge inconsistency for mechanical workpiece by machine vision with deep learning and variable geometry model

Inconsistent edges of mechanical workpieces in the same batch are one of the main reasons that lead to different machining performance. A full recognition of the inconsistent features has significant impact on enhancement of their intelligent machining ability. An intelligent hybrid strategy is proposed for edge inconsistent feature detection by machine vision, in which deep learning is combined with variable geometric model together to conduct the function. A deep convolutional neural network based feature classification model is established with K-Means clustering tactic. Supported on the classification model, a variable geometric model for specific edge inconsistency is given with an inconsistency evaluation function to investigate the match degree between the geometric model and the actual detected edge, and then particle swarm optimization algorithm is applied to find the solution of this geometric model. Detection experiments are carried out on a domestic servo-driven vision measuring platform to verify the performance of the proposed approach. The results show that the combined scheme can classify the different type of geometric contour of edge features with 100% correctness, and better evaluation performance by dice similarity index and Hausdorff distance in comparisons with other recent candidate methodologies. It is also indicated that the presented method provides a good recognition of the geometrical shape with less than 0.06 m m maximum error for workpiece with 142 × 119 m m size in the visual field.