Design Of Process Recognition System Of Magnetic Material Smelting Furnace Based On Machine Vision

The production process of sintered Nd Fe B magnets and alloy magnetic materials con-tains several production processes,and each process plays a vital role in the production of magnetic materials,so the inspection of its process is one of the most important tasks in the production process of magnetic materials.In this paper,we propose a process recognition solution based on machine vision and machine learning,and design an image-based process recognition method,which combines target detection algorithm and decision tree classifica-tion method to identify the production processes of magnetic materials.The production process of magnetic material mainly includes 8 processes before the furnace and 8 processes at the top of the furnace.Before the furnace,the furnace door is opened,melting,discharging,polishing copper rollers,personnel observation,wiping the furnace chamber,wiping the furnace door and closing the furnace door? at the top of the furnace,the furnace lid is opened,slag cleaning,dosing,charging,wiping the furnace mouth,pouring crucible,melting and closing the furnace lid.In this project,16,000 images of the furnace front and furnace top processes are selected,and a target detection dataset of the furnace containing target objects such as people,equipment and tools is made,and the YOLO v5 target detection model with high detection accuracy and fast calculation speed in the target detection algorithm is used to detect the relevant targets.By setting the depth of the decision tree and pruning according to the minimum number of samples of the leaf nodes of the decision tree,the decision tree classification model is designed to determine the current production process by the target in the image and its relative position information.Finally,the system interface is designed by QT Designer,and the process recognition system is built based on Py Qt5 framework and YOLO v5 target detection algorithm.After the experimental test,the m AP of YOLO v5 target detection algorithm is 95.0%,the average detection accuracy of the process recognition in front of the furnace is 96.4%,and the average detection accuracy of the process at the top of the furnace is 95.0%,which basi-cally meets the process recognition requirements of the magnetic material melting furnace,and also provides a useful attempt for the application of artificial intelligence technology in the field of magnetic material production.