Research On Lidar SLAM Technology For Substation Inspection Robot

In recent years,China vigorously develops smart power grid,the scale of substation gradually expanded,and the functions of equipment in the station increased.As a result,the work intensity of substation inspection is constantly increasing,and the professional requirements of inspection become ever higher.The unmanned inspection robot of substation has become an important means to solve the problem.Simultaneous Localization and Mapping(SLAM)is a key technology for robot inspection.Currently,lidar SLAM autonomous mobile robots have been widely deployed in simple indoor scenes such as hotels,shopping malls and offices.However,SLAM technology in complex outdoor operation and maintenance scenes of substations is still immature,which has attracted great research interest and development.This paper carried out an in-depth study on substation inspection SLAM technology,pointed out the limitations of the current application of SLAM technology in the substation scene,and put forward a targeted multi-sensor fusion lidar SLAM algorithm to solve these problems,which can effectively improve the positioning accuracy and global mapping quality in the substation scene.The main research works are as follows:1)A two-dimensional lidar SLAM system is designed,which uses the fusion data of IMU and wheel speed meter as the initial matching value,and uses the improved resampling strategy of RBPF(Rao-Blackwellized Particle Filters)as the front motion-odometer to effectively solve the problem of particle dissipation.At the same time,the multi-submap system adopts special maintenance mechanism and accelerated loop search strategy to provide more effective odometer constraints and faster loop closure detection.In addition,the backend optimization module based on graph optimization is added to the RBPF filter,which greatly improves the fast error divergence of particle filter SLAM and the global consistency of the system.2)A 3D point cloud preprocessing method for complex operation and maintenance scenes of substation is designed.The method consists of three modules: point cloud motion distortion removal,ground extraction and point cloud cluster grouping.The point cloud motion distortion removal algorithm uses a method to estimate the velocity between laser points.The sampling velocity and relative pose of each laser point are calculated point by point to recover the undistorted coordinates of each laser point.Ground point cloud extraction adopts a special ground point extraction strategy based on ground plane fitting,which is more in line with the ground physical model and can effectively adapt to complex ground conditions.Finally,a fast-clustering method for point cloud of large equipment is designed in the substation environment,which can extract and classify point cloud clusters of large equipment at the fastest speed while ensuring the correct classification.3)The NIMLS-ICP(Normal and Implicit Moving Least Squares – Iterative Closest Point)algorithm is proposed,which is more consistent with the characteristics of point cloud.The normal vector and curvature constraint are used to screen the upcoming matches,so as to reduce the probability of false matches and reduce the number of iterations.In addition,the ICP error equation is constructed by using the moving implicit surface.The constraints are constructed jointly with the normal vector to carry out the least square iterative calculation of error,which makes the error model more in line with the real world and improves the matching accuracy of the algorithm.