Research On Visual Positioning And Classification Method Of Cylinder Head Automatic Sorting Robot Based On Deep Learning

Intelligent manufacturing is one of the most important industries in the "Industry4.0" era,and it is also an important supporting technology field of the "Made in China2025" plan.The booming development of artificial intelligence has given new energy to smart factories.At a time when the COVID-19 epidemic has seriously affected human life and production,it is more meaningful to use robots instead of humans to complete various complex and labor-intensive jobs.Engine cylinder head is one of the key parts of the automobile.At present,the production of automobile engine cylinder head is mostly carried out by manual sorting and handling,which has low efficiency and certain hidden dangers for safe production.In the face of complex actual requirements,it is of great engineering and academic significance to apply the new theory and technology of artificial intelligence to the cutting material sorting process of engine cylinder head production,to locate and classify cylinder heads by visual method,and to guide the robot to carry out intelligent sorting of cylinder heads.Therefore,a deep learning-based engine cylinder head classification and positioning network model is proposed in this paper.Then,in order to further position the engine cylinder head hole with higher precision,a high-precision positioning method of the cylinder head hole circle based on edge search and fitting was proposed.The method presented in this paper has excellent performance in both speed and precision.The main work of this paper is as follows:1.Collected and produced cylinder head data set.The cylinder head pictures collected in the process of on-site production and off-line material were marked and data enhanced.Used to train engine cylinder head positioning and classification network.2.The visual localization and classification of convolutional neural network are proposed.The residual network structure with lightweight design is used as the backbone network,which greatly reduces the computational complexity of the network structure.At the same time,attention mechanism is introduced in multi-scale detection.Finally,a loss function is designed based on EIOU to improve the accuracy of positioning.3.A comprehensive circle detection method is proposed,which greatly improves the accuracy of circle detection on the premise of ensuring less impact of real-time detection.4.The imaging device of cylinder head intelligent sorting off line identification system was designed,including the selection of camera and light source,lighting mode and Angle.A complete set of vision system software was developed,including calibration,identification,classification,communication with PLC and robot,etc.,and the proposed method was successfully deployed in actual production.The proposed visual positioning and recognition method of engine cylinder head cutting robot realizes the high speed,high precision positioning and accurate classification and recognition of engine cylinder head targets.The proposed method was tested on a test set composed of 530 cylinder head images.The location accuracy of the proposed method was 95.9%,the location error of the hole circle was less than 9 pixels,the classification accuracy was 98.5%,and the detection speed was 8ms.