Threaded hole detection method based on guided filtering and neural network algorithm

In order to solve the problems of tedious operation, poor accuracy and low efficiency in measuring the screw hole of crankshaft end face, a screw hole detection method based on guided filtering and neural network algorithm was proposed, taking the model YC4W75 crankshaft as an example. Firstly, real-time grabbed images were preprocessed using guided filtering and morphology to eliminate the effects of surface noise and mottle. The edge features of the internal thread path in the crankshaft end face were extracted. Then, combined with RANSAC algorithm, a neural network model was constructed using Pytorch to fit the extracted circle. The size of the internal thread path on the crankshaft end face and the distance between the centers of the circle were obtained. The test results show that the error of the small path of the internal thread is less than 0.070 mm, and the error of each thread hole and the center hole is less than 0.200 mm. The proposed method can meet the accuracy requirements of the industrial site due to high measurement accuracy and the operation simplicity. The automatic measurement of the position information of the threaded hole of the crankshaft end face is realized.