Research On Detecting Method Of Aviation Bearing Assembly Ball

Ensuring the quality of aviation bearings is very important for aviation safety.As precision bearings,aviation bearing components have strict dimensional error requirements.In order to guarantee the quality and performance of the factory bearings,it’s critical and necessary to inspect the assembled bearings before they leave the factory.This article aims to detect the qualified size of the aviation bearing ball.For the assembled bearing,it can be seen that the gap between the inner and outer ring of the ball and the cage is narrow,the visible part of the ball is small,the light is weak in the gap,and the local edge of the ball is short.It is difficult to realize detection difficulties such as the difficulty of arc circle fitting,as well as highprecision detection requirements.A visual inspection system is designed to detect and measure the balls of the assembled bearings.This paper designs and builds an aviation bearing ball assembly inspection system platform,which consists of a hardware module suitable for local image acquisition of bearings consisting of a camera,a light source and a computer,and a software image processing module for ball inspection of the acquired images.First of all,the suitable light source environment is designed for the problems such as serious reflection on the bearing surface and insufficient light on the gap between rings and cage,and the camera lens model is selected according to the demand of high accuracy of inspection.Secondly,an aeronautical bearing ball detection program including image acquisition,image preprocessing,semantic segmentation of ball targets,ball edge extraction,edge circle fitting,and ball measurement based on camera calibration is designed.When segmenting to obtain the ball features,in order to solve the problems of the target being occluded,the target and the background gray level contrast is not obvious,and the amount of trainable image data is small,the segmentation edge requirements are fine,etc.,a segmentation model is designed which based on the U-Net+++ model.The model’s underlying convolution structure is improved to improve the information loss problem caused by pooling of the original network by cascading the null convolution,and replacing the lowdimensional activation function to avoid The Re LU function in low dimension causes information loss.Then,edge extraction is performed on the local features of the ball obtained by segmentation,and the size of the ball is measured by the circle fitting algorithm based on RANSAC combined with the camera calibration to judge whether the ball is qualified.Finally,this paper verifies the above algorithm through experiments.The results show that the MIo U of the U-Net+++ based aerospace bearing ball segmentation model designed in this paper reached 90.12%.The system achieves an accuracy of0.001 mm for aerospace bearing ball inspection,with an average inspection error of about 4μm and an error detection rate of only 1.3%.The method in this paper realizes the detection of the embedded ball of the assembled bearing,and provides a method and idea for the detection of the ball in the inspection of the bearing assembly.