Research And Design Of Taper Roller Appearance Defect Detection System Based On Machine Vision

With the development of computer technology,modern industry is becoming more and more important to the degree of automation,and the industrial system is developing towards the intelligentization,which makes the surface quality of industrial products more and more demanding.The traditional surface quality inspection mostly through the way of visual inspection,not only inefficient but also poor stability.As a result,machine vision-based inspection technology has evolved and flourished.In this paper,we focus on the appearance defects of tapered roller,using visual inspection technology and image processing and analysis methods to develop a complete visual inspection system,including hardware deployment and software implementation,and designed a targeted surface defect detection Image processing algorithms,a set of effective surface defect detection solutions are proposed.The development tools of this paper are based on C # and NI Vision platform.The development and programming are efficient,and the development time is greatly shortened.The image processing algorithms are designed by OpenCV and MATLAB for upper software call.This paper discusses the common light source scheme and other visual detection hardware,and designs an optical imaging system suitable for roller detection with multi-angle illumination for image acquisition.It compares the common edge detection algorithms and threshold segmentation in several image segmentation The theory of feature extraction and the common pattern recognition classifier are introduced.Aiming at defects such as damage,lacking of grinding,rust and lack of chamfer,a set of effective image processing algorithm flow is designed,and the maximum intraclass variance method and local threshold algorithm are proposed to segment the defect,and feature extraction is realized for the segmented image And description,and then SVM classification and identification.The application of the system in actual production shows that the system can realize the appearance defect segmentation and recognition of the tapered roller.This paper also introduces the development of depth learning and its application in target detection.The structure and realization of several commonly used convolutional neural networks are described in detail,and the improvement and optimization of the target detection framework are analyzed.A pre-training model based convolution neural network is designed by using the method of migration learning,and the network fine-tuning is carried out for the two defect target data sets.The training speed of the model is quick and the recognition effect is remarkable,which is superior to the traditional defect detection algorithm.This paper realizes the on-line detection of the appearance defects of the tapered roller,and the detection results are efficient,reliable and robust,and can replace the artificial visual effect.The results of system operation show that the algorithm has high accuracy and good adaptability,which can meet the actual demand in industrial production.