Research On Riveting Quality Detection Method Based On Machine Vision

Press riveting is one of the important ways of riveting connection,which plays an irreplaceable role in automobile assembly,rail transportation,aerospace and other high-end equipment manufacturing fields.Therefore,the qualified riveting quality is an important guarantee for the safe and stable operation of the equipment.As an important indicator for measuring the quality of riveting,the appearance of rivet heads can directly reflect the quality of riveting.At present,the inspection of the appearance of rivet heads is mainly based on manual inspection,which is greatly affected by subjective factors,has low detection efficiency,and can easily lead to missed or incorrect inspection.This paper designs an online detection system for riveting quality and conducts research on riveting quality detection methods.This method meets the automated assembly production requirements of multi rivet riveted parts and improves the inspection efficiency and accuracy of riveting quality.The main research contents are as follows:1.A riveted quality inspection method based on feature fusion and ELM is proposed for the existing riveted defect detection methods,which have problems such as difficult localization and low detection efficiency.This method completes the detection of riveting quality for pressure riveted parts with uniform distribution of riveting points.Firstly,Bilateral filter and improved Retinex image enhancement algorithm are used to preprocess the rivet point image,which eliminates the effects of noise,scratches and reflection.Secondly,a multi threshold object segmentation algorithm is proposed to obtain regions of interest such as rivets and upsets.Then the multi-dimensional features such as perimeter,area,roundness rate,eccentricity rate and Zernike moment of the region of interest are extracted.And perform fusion and normalization operations on image features.Finally,a classification algorithm based on ELM was designed to achieve the classification of riveting defects.Through experimental verification,this method can meet the real-time online detection requirements of riveting defects,with high detection accuracy compared to traditional SVM and BP detection algorithms.The correct detection rate of "qualified-defect" binary classification and defect multi classification are 95.2% and 92%,respectively,which have good overall detection performance.2.A riveting quality detection method based on improved DETR is proposed to solve the problems of unclear morphological features of rivets and rivet heads,and difficulty in segmenting targets under complex working conditions such as small rivet sizes,uneven distribution of riveting points,and uneven riveting surfaces.This method further improves the detection accuracy of riveting defects.Firstly,EfficientNet is used as the backbone feature extraction network in DETR,and the 3-D weighted attention mechanism SimAM is introduced into this network.It is able to effectively retain the header morphological information and the spatial information of the rivet region in the image feature layer.Secondly,a weighted bidirectional feature fusion module is introduced in the DETR framework,with the output of the EfficientNet network as the input of the fusion module.It aggregates multi-scale feature information and increases the inter class differences of different defects.Finally,the regression loss function of the DETR prediction network is improved by using the linear combination of Smooth L1 and DIoU,which improves the convergence speed and detection accuracy of the model.The experimental results show that the improved model exhibits high detection performance,with an average detection accuracy mAP of 97.12% and a detection speed FPS of 25.4 f/s for riveting defects.Compared with other mainstream detection models such as Faster RCNN and YOLOX,the improved model has significant advantages in detection accuracy and detection speed.This method achieves rapid identification and precise positioning of riveting defects.