Study On The Dynamic Recognition For The High Speed Rail Furnace Number Before Welding

In order to achieve the goal of quality control,process management and product trace-ability during high-speed rail welding process,it is necessary to input the character information,such as rail batch number and rail body number rolled on the rail surface,into the ERP system used in the workshop.The current solution adopted by the factory is in a manual way,which is cumbersome to operate,labor-intensive and error-prone.Using automatic character recognition technology is conducive to efficiency improvement and the labor intensity reduction for operators.Under the complicated production conditions of the rail welding workshop,the dynamic recognition of the rail batch number is more difficult than the ordinary optical character recognition.Aiming to overcome the problem caused by environmental effects and rail self-defects during the dynamic recognition process,an automatic system for the recognition of high-speed rail batch character and its related hardware structure based on 3D laser sensor were proposed in this paper.MVTEC HALCON image processing interface combined with C# programming was adopted to complete the image reconstruction algorithm design and character recognition software development,which realized online dynamic recognition process.The main contents of this paper are as follow:(1)To solve the problems of high-speed rails with self-defacement,difficulty in character image extraction and real-time image acquisition,3D laser sensors were selected as image acquisition devices.The outline reconstruction from point cloud information about rail was completed by data splicing and the image including the rail batch number characters information was obtained by the method of shape matching.(2)HALCON image processing operators were used as the core of the image pre-processing.And different image pre-processing operators such as mean filter,variable structure element median filter,Gaussian filter,threshold segmentation,variable structure element morphology processing and regional feature extraction were studied in the paper.The different algorithms of character image processing were studied to achieve the best image pre-processing steps and methods,and build up the algorithm design of the character image processing finally.(3)Multiple neural network(MLP)algorithm was used as character recognition algorithm by constructing several different recognition classifiers.The accuracy and confidence coefficient of the character recognition were improved by multiple normalized character features extracting with the back-propagation training from rail batch number library.(4)The hybrid programming between HALCON and C# language was accomplished by analyzing the language structure of C# and HALCON.And the software for the rail batch number characters recognition was developed.The actual operation results indicated that the system could run reliably in the rail production field,and the one-time character recognition rate could be up to 95%.