Research On Displacement And Vibration Measurement Technology Via High-Speed Vision System

With the rapid development of economic construction and the continuous improvement of science and technology,higher requirements are put forward for the performance of industrial equipment and the precision of products,especially in the field of measurement technology.It is necessary to measure position,displacement and vibration in electronic industry,automobile industry and medical imaging,and the commonly used measurement methods are contact measurement,such as position sensor measurement,wire-drawing displacement meter measurement,GPS satellite measurement,etc.With the improvement of basic hardware facilities,non-contact measurement has gradually begun to be applied,such as laser,ultrasound and vision measurement technology,especially Vision measurement technology,can keep space from the object.First,the image of object's movement is collected by the camera,then the accurate displacement can be obtained with computer data analysis.And not only the static objects,but also dynamic objects can be tracked and measured.Machine vision,as an innovative technology,can meet the demand of high precision automation production in domestic industry.In this paper,a method of measuring displacement and vibration based on high-speed vision is studied and implemented by analyzing and summarizing some shortcomings in measuring displacement and vibration in engineering field.The main research contents and results are as follows:(1)The specific structure of the high-speed vision measurement system is studied,and the light source,lens,processor and control system are analyzed and introduced respectively,and the specific parameters of the hardware used in this paper are defined.A camera measurement model is established,and its internal and external parameter matrix is calibrated.A coordinate conversion method of sampling points in high-speed vision measurement system is proposed.(2)As there is always the problem of Perpendicularity Error between optical axis and measuring plane in two-dimensional vision measurement,which it is difficult to be calibrated,a method of oblique optical axis correction based on BP neural network is proposed.it is concluded that internal and external parameters of camera are the main factors for affecting measurement accuracy,by theoretical analysis of measurement error caused by oblique optical axis of camera in two-dimensional vision measurement.Based on this,the feasibility of oblique optical axis correction using BP neural network is analyzed.A visual measurement platform is built.After collecting pictures,data sets are pre-processed,and sigmoid function is used as activation function,BP neural network model is constructed,and data and labels are imported for training.The training results show that the difference between theoretical value of the neural network training model and the actual value is small,and the actual deviation angle can be calculated effectively by deducing the acquired pixels.By calculating LOSS deviation value,it is found that the fitting effect is well,the deviation is small.A validation test is designed based on the built visual measurement platform.Tri-axis test turntable is used as the test platform,the theoretical calculation data,and neural network fitting and real data before and after correction are compared.(3)By extracting a series of points which have the greatest impact on mechanical motion,an analysis algorithm is designed for the mechanical vibration characteristics,which provides an advanced measurement technology with non-destructive,convenient and visualization.This algorithm is applied in practice,and good results are obtained,which proves the reliability and robustness of the algorithm.(4)A displacement measurement algorithm based on Kalman filter and local template matching is proposed.The state equation and observation equation of the moving target are established.The position of the moving target in the next image is predicted continuously with the recursive relationship.The specific position information of the target is detected by the local template matching algorithm to improve the search efficiency.Meanwhile,it can also solve the problem of losing a moving target in photography by high-speed camera.The experimental results show that the average absolute error of the method is1.3395×10^-2mm,and and mean relative error is 1.75×10^-2mm.After adding Kalman filter to the displacement measurement algorithm of feature matching,the accuracy of point recognition can be improved.when the occlusion area is less than10*20 cm²,the accuracy can reach 99.87%.When the occlusion area is greater than10*30 cm²,the accuracy can not match.