Efficient Adaptive Surface Measurement Based on Multilevel Gaussian Process Modeling and Scanning Path Optimization

Abstract

Adaptive surface measurement based on Gaussian process (GP) reduces data redundancy while maintaining the same level of surface reconstruction accuracy as compared with conventional raster scanning. However, the relatively long scanning distance and GP computation time may limit its application when scanning also accounts for the time cost. Toward the simultaneous comprehensive optimization of sampling amount, scanning distance and measurement efficiency, a strategy combining two-level GP modeling and ant colony path planning is proposed. Faster local GP guided sampling along the scanning path is nested to reduce global GP modeling iterations. Experimental results validate that the time-consuming global iterations can be reduced to ~10% of the common GP adaptive method. Compared with raster scanning, without sacrificing the accuracy of surface reconstruction, the sampled data amount, total scanning path length and time cost for measuring a free-form surface are reduced to about 9%, 22% and 13%, respectively. Owing to these prominent advantages, the proposed method has potential applications in efficient and accurate surface measurements.