Design And Implementation Of Surface Quality Inspection System For Complex Parts Based On Machine Vision

Manufacturing industry is the lifeblood of national economy.Now China and China are entering the ranks of major manufacturing countries,but it is not a powerful manufacturing country.Its outstanding problems are mainly product quality problems and low labor productivity.In order to improve our core competitiveness,intelligent manufacturing should be the main direction of attack.In intelligent manufacturing,machine vision technology is considered as an important part and the most efficient tool to improve the automation of production links.This paper focuses on the surface quality detection of complex parts and applies machine vision technology to study the quality detection of mechanical parts from five aspects: image acquisition,image preprocessing,feature extraction,detection and analysis,and conformity sorting.The purpose is to put forward a quality detection system with strong universality,rich functions,high accuracy and fast speed.The main contents of this paper are as follows:(1)Design the hardware system of machine vision inspection.The industrial cameras,lenses,light sources,mechanical arms and controllers used in machine visual inspection are selected according to the requirements of inspection,and positioning elements are designed to ensure that the position of the detected parts in the field of view is relatively fixed.Meanwhile,the cameras and light source supports are designed to realize multi-direction adjustment of the detection system according to the characteristics of the tested parts.Finally,the production line system is arranged reasonably to ensure that the relative position of the mechanical arm and the conveyor belt is fixed in the on-line inspection.(2)Design modular frame vision algorithm.Making some commonly used image processing algorithms,such as image acquisition,image preprocessing,image segmentation,feature extraction,report writing,login interface,etc.,into the form of sub-programs,can realize fast integration and test of new algorithm.Only by simple modification of configuration parameters and combination of each module,various complex visual inspection tasks can be completed,the time of field debugging can be compressed,and the difficulty of technical verification can be reduced.(3)Construct defect type database.For common scratch and speckle defects,a database was made.200 images were collected for each type of defect on the inner and outer surface of the parts to be detected.With common data expansion method,800 pieces of each type of defect were expanded to improve the robustness of the training model.(4)Improve the quality detection method.For dimension measurement,abandon the traditional single template calculation method,extend the detection path to two-dimensional space and only detect a specific area,reduce the amount of calculation and improve the real-time performance;For defect detection,traditional defect detection algorithm based on threshold segmentation and pattern matching and SSD＿based on deep learning are used.Mobilenet V2 defect detection algorithm is compared and tested to analyze their respective application scenarios.At the same time,a one-button deep learning training module is designed to simplify the training process.(5)Optimize the reasoning process of in-depth learning.The disadvantage of traditional reasoning using GPU is analyzed,and a reasoning model based on CPU+Open VINO is put forward,which reduces the reasoning time to one-sixth of the original one and greatly improves the detection efficiency.(6)Develop quality inspection system for mechanical parts.The module is built by two working modes,offline test and online test.In offline testing,image acquisition module,image preprocessing module,feature extraction module,dimension measurement module,defect detection module and report writing module are mainly used to complete high-precision multi-parameter and multi-surface detection of parts.In online detection,image acquisition module,dimension measurement module,defect detection module and communication module are mainly used to realize high-speed and parallel detection by deploying acquisition and detection programs to CRIO.