Research On Defect Detection Of Injection Molding Products Based On Machine Vision

In the continuous production process of injection molding products,it is inevitable to produce defective products.In order to improve the product yield,it is necessary to pick out the defective products.The traditional method is to rely on manual inspection.In addition,in the process of building detection models,traditional machine vision algorithms rely too much on skilled experience to build models,and the models are not universally applicable,resulting in high installation and deployment costs of traditional visual detection methods in enterprises,which is not conducive to later maintenance.The visual detection method based on neural network can complete the model construction more efficiently.This paper proposes a defect detection system based on neural network,which can not only realize the task of defect classification,but also locate the defect position.The main research work are as follows:(1)According to the characteristics of injection molding products,a set of defect detection hardware devices was designed to complete the image acquisition task,the selection of important components and the arrangement of light sources were completed,and a diffuse reflection dome lighting detection platform was built to solve the problem of high reflectivity and less prominent defects of products.(2)According to the analysis of the original data set,an improved model and preprocessing scheme based on YOLOv5 s are proposed.Based on the analysis of different data preprocessing effects,the structure of the YOLOv5 s model and the evaluation method of defect target detection are introduced,and the multi-scale defect detection performance of the YOLOv5 s model is improved for the problem of poor detection of small-size defects,and the m AP is increased by 2.18%.(3)According to the analysis of injection molding production requirements,a defect detection human-computer interaction software was designed.The user login registration module and the defect detection module are designed respectively.Then the comprehensive test of the defect target detection system for injection molded products shows that the detection model can adapt to certain environmental disturbances.The real-time detection accuracy of defective products is 83.2%,the FPS reaches 22.17,and the missed detection rate of defective products reaches 1.87%,it basically meets the high frame rate realtime industrial injection molding surface defect target detection task.