Genetic Algorithms and Machine Learning Enhanced Laser Displacement Sensor Point Cloud Augmentation for Low-Cost, Large-Scale Flatness Measurements

The accurate measurement of the flatness of satellite devices is of great significance for the performance of satellites; however, the planes to be measured on satellites have different sizes. Using a limited number of sensors to measure the flatness of large-scale planes, the point cloud repositioning required for multiple samplings will cause a sharp increase in the computation. The reference planes are inconsistent in multiple samplings. To solve these problems, this article proposes a low-cost, large-scale flatness measurement method based on genetic algorithm (GA) and machine learning. First, this article proposes a regional measurement method based on a rotating platform and a laser sensor, which realizes point cloud augmentation, adapts to different scales and shapes of planes, and reduces equipment costs. Second, a point cloud correction method is proposed, using GAs and spatial transformation algorithms to reduce the impact of mechanical tolerance in the rotating platform on measurement accuracy and enhance measurement result consistency. Finally, this article uses a machine learning algorithm based on the gradient boosting decision tree (GBDT) model to perform flatness error regression, which reduces the influence of nonlinear errors in data on measurement results and simplifies and accelerates the measurement process of flatness. The experimental results show that the proposed method can achieve efficient, accurate, and nondestructive flatness error measurement, reducing the relative error of flatness measurement of large-scale planes from 52.7% to 0.05%, enabling the use of the same equipment for planes of different scales.