Development Of Grating Displacement Measurement System Based On Neural Network And Machine Vision

In the process of manufacturing and assembling these components,the requirement of processing accuracy in many fields such as integrated circuits,ultra-high precision processing,precision measuring instruments and so on is constantly increasing,and the requirement of precision measurement accuracy for corresponding micro-electronic manufacturing equipment is also increasing.Therefore,higher speed and higher precision are required for displacement measurement system of microelectronics manufacturing equipment.This paper presents an innovative raster ruler image measurement system based on neural network model.The system can effectively overcome the limitation of traditional grating measurement accuracy(usually 20um)and greatly improve the operation speed of measurement feedback through neural network model algorithm.The results show that the accuracy of displacement measurement is higher with lower resolution in precision measurement.Firstly,the research status of grating rulers at home and abroad is analyzed from two aspects: the industrial application status of grating rulers and the research status of grating rulers.There are contradictions between high speed and high precision in grating scale displacement measurement system,and nanometer resolution.The highest accuracy can only reach micron level,that is,there is a big error between resolution and measurement accuracy.In view of the above problems,this paper,based on the theory of neural network,uses digital image technology to eliminate the manufacturing error of precision displacement measurement system and the mathematical modeling error of non-linear system.Secondly,due to the data set pictures collected by industrial cameras,they are vulnerable to environmental noise.So the raster image processing technology is used to process the data set,and the data set is made according to the input form requirements of the neural network model.In this paper,we propose a new image processing method to judge the position of grating and establish the corresponding algorithm based on this method.Then,we use the neural network model to improve the traditional algorithm modeling based on physical knowledge.Because the input data of the neural network model is the characteristic window of the grating ruler image and the output is the only mark of each grating,according to the actual situation of the precision displacement measurement system and according to the theory of the neural network,the appropriate neural network model and the corresponding algorithm are selected.Furthermore,according to the number of gratings and characteristic image windows and data sets,the topological structure,number of nodes and parameters of the neural network are selected.Finally,through Python development environment and TensorFlow framework,a neural network model is built to identify each grating.Then the displacement of the moving platform is calculated by the displacement measurement program and compared with the displacement data of the interferometer.The experimental results show that the recognition rate of each grating based on the neural network model is over 99%.And the accuracy of displacement measurement is up to±1um.By comparing the experiments,we find that the new image measurement system not only improves the resolution,but also greatly improves the accuracy of displacement measurement.It also significantly reduces the manufacturing accuracy requirements of grating ruler,and effectively reduces the manufacturing cost.