Adaptive Sampling Strategy Based On Gaussian Process

Atomic force microscopy(AFM)is one of the most frequently used instruments in the field of micro and nano measurement.The main function is to measure the morphological and mechanical characteristics of the samples.At present,with the increasing demand for measurement efficiency and accuracy,a new adaptive sampling strategy research trend of high efficiency and precision is emerging.Machine learning is a science of artificial intelligence which can predict unknown behavior by learning existing data or experience.Among them,Gaussian process is a powerful model that can model and predict data with a non-parametric mechanism.It plays a very important role in machine learning and the advantage is that it can fit the black box function and give prediction uncertainty.In this paper,on the premise of ensuring the measurement accuracy and improving the measurement efficiency,the intelligent adaptive sampling and automatic contour tracking based on Gaussian process are studied in order to improve the efficiency of surface topography measurement of nanometer measuring instruments such as atomic force microscopy.There are two main aspects of the work:1?Scanning path optimization combined with adaptive sampling strategy.In this part,we use two scanning paths,Archimedes helix and Hilbert curve,and combine adaptive iteration algorithm based on Gaussian process to improve measurement efficiency.Sampling points can be greatly reduced by intelligently selecting sampling points along scanning paths and reconstructing sample morphology based on Gaussian process.On the premise of ensuring accuracy,uniform sampling reconstruction on the path can reduce sampling points to about 12%of raster sampling points.On the path,the number of points can be reduced to about 4%after the reconstruction of the intelligent sampling points.It can be seen that the measurement time will be greatly reduced and the efficiency will be greatly improved.The total contact time between the probe and the sample will also be reduced,and the wear of the probe will be reduced,which can improve the life of the probe.2?Curve tracking measurement.When the measured object is a linear object,such as nanowires,DNA,boundary lines,etc.There will be a large number of non-characteristic areas on the substrate that need not be measured.Raster scanning will measure the entire substrate,which will obviously cause a lot of data redundancy.In order to reduce this unnecessary waste,a curve tracking measurement strategy based on Gaussian process is studied.This strategy tracks the outline of interest,and the substrate without sample characteristics is not measured.Simulation and experimental verification show that it can reduce the scanning path length of an order of magnitude.Obviously,it can reduce the number of sampling points and improve the measurement efficiency.This method has great potential for application in atomic force microscopy.