Research On The Correctness X-ray Detection Algorithm Of Complex Structural Parts Based On Deep Learning

In the production process of related products such as aerospace industry,national defense military,and automotive high-speed rail in China.There are a large number of important products with various internal parts and complex assembly structures.In order to ensure the stability of each component in the production process,it is one of the indispensable processes to detect the correctness of the assembly of its internal components.Although the previous detection method,which is based on extracting and matching characteristics such as the shape of the connected regions in sample images,the aspect ratio or the area,achieves a better detection result.But by mechanical precision,assembly tolerances,parts dislocation and other factors,it is less robust.To this end,we study the problem of the correctness of the assembly of internal parts of complex structures.Comprehensive X-ray multi-view imaging principle and convolutional neural network algorithm,Automatically identify the region of interest in the image.So that the criteria for the determination of qualified products from the regional characteristics into individual characteristics.First of all,according to the task requirements for the detection of the correctness of the internal parts assembly of complex structures,design the overall scheme.Using the X-ray imaging system to collect the projected sequences of different angles as the data support of the deep learning network,the different parts in the workpiece are marked as different types.After that,design a deep CNN model to train and classify the internal parts,and output the coordinate of every part.For the first-time detection of the workpiece that is judged to be a defective product,perform a secondary detection to avoid misdetection.For workpieces that have completed missing detection,we find standard workpiece views that conform to the current test-workpiece’s angle based on the sine properties of the projected data distribution.Perform assembly error detection,such as transposition or dislocation of the test workpiece’s internal parts.Finally,for the stability and reliability requirements of the detection system for the correctness of the assembly of complex structures.We carried out experiments and analysis on two kinds of simulation workpiece products and actual products respectively,and completed the identification of missing and transposition of internal parts of the workpiece.And using a variety of assistive technologies to improve the target recognition accuracy of the detection algorithm,the accuracy of the whole system to the internal parts of the workpiece is not less than 90%.The whole detection process is fast and efficient,and it has certain robustness to the mutual covering of its internal parts,which meets the requirements of the system.