automatic intelligent inspection system for crankshaft grade detection based on ‎machine vision and deep learning

The adaption of main bearings with crankshaft grades is an important consideration in bearing installation tasks. if an operator is not careful, it will cause a significant decrease in the quality of the final assembled engine and also cause some defects. machine vision systems have the potential to implement autonomous error detection that can significantly reduce inspection time and lead to more frequent, precise, and objective inspections. herein, an inspection system was developed, capable of automatically detecting crankshaft grades from crankshaft images.  a specific lighting condition was designed to obtain proper images of the crankshafts. an efficient diagnostic approach based on the semantic segmentation method was presented in this regard. two different convolutional neural network (cnn) architectures, including mobilenet and vgg19, were trained and evaluated. mobilenet was revealed to be the best compromise between accuracy, with an iou-score of 85%, and validation time, with 0.2 ms for discovering the characters engraved on the crankshaft. according to the obtained results, the proposed approach could be used as an efficient, accurate, and fast tool for the automatic detection of crankshaft grades in bearing assembly stations.

automatic intelligent inspection system for crankshaft grade detection based on ‎machine vision and deep learning

the adaption of main bearings with crankshaft grades is an important consideration in bearing installation tasks. if an operator is not careful, it will cause a significant decrease in the quality of the final assembled engine and also cause some defects. machine vision systems have the potential to implement autonomous error detection that can significantly reduce inspection time and lead to more frequent, precise, and objective inspections. herein, an inspection system was developed, capable of automatically detecting crankshaft grades from crankshaft images.  a specific lighting condition was designed to obtain proper images of the crankshafts. an efficient diagnostic approach based on the semantic segmentation method was presented in this regard. two different convolutional neural network (cnn) architectures, including mobilenet and vgg19, were trained and evaluated. mobilenet was revealed to be the best compromise between accuracy, with an iou-score of 85%, and validation time, with 0.2 ms for discovering the characters engraved on the crankshaft. according to the obtained results, the proposed approach could be used as an efficient, accurate, and fast tool for the automatic detection of crankshaft grades in bearing assembly stations.