Research On Visual Navigation Method For Substation Inspection Robot

With the continuous development and progress of the power industry,there is an increasing focus on the expansion of the power grid scale and the renovation of aging power grids in many countries.To ensure the safe and reliable operation of power systems,daily maintenance of the power system is particularly important.Traditional manual inspection methods are not only inefficient,but also face issues such as insufficient number of operation and maintenance personnel,and differences in skill levels,making it difficult to adapt to the inspection requirements of modern power grids.Therefore,developing an intelligent inspection device that can replace manual inspection is an inevitable trend.In recent years,with the rapid development of artificial intelligence(AI)technology,inspection robots have been applied in various industrial fields,and using inspection robots to replace manual labor for complex and hazardous inspection tasks has become an important means to improve inspection efficiency and reduce labor costs.To enhance the safety of the robot during the inspection process,this thesis proposes an improved obstacle detection model based on YOLOv5 for substation inspection robots,as well as a path planning algorithm that integrates obstacle category information with laser sensor information.The specific work is as follows:(1)Analyzing the current environment of substations and the requirements of substation inspection robots,we built a substation inspection robot model based on the ROS system.Convolutional neural networks(CNNs)are introduced as the theoretical basis for object detection models,and through experimental comparative analysis of mainstream object detection models,YOLOv5 is selected as the base model for obstacle detection in this thesis.(2)Addressing the issues of large parameter size,high computation complexity,and the tradeoff between detection accuracy and real-time performance in existing object detection networks,we propose replacing the original CSPDark Net backbone structure of YOLOv5 with the lightweight Ghost Net network structure.We further enhance the feature extraction capability of the obstacle detection model by introducing the CBAM attention mechanism into the feature extraction and fusion parts of the model.Experimental results show that the improved YOLOv5 model can improve real-time performance while maintaining accuracy.(3)Through comparative analysis of global path planning algorithms,we propose an improved algorithm based on the D* algorithm.We address the issues of sharp corners and poor smoothness in the original algorithm by introducing B-spline curves for improvement,and the effectiveness of the proposed method is verified through simulation results on the Matlab platform.(4)After comparing and analyzing two local path planning algorithms,DWA and TEB,we select TEB as the base local path planning algorithm for this paper.We also integrate the proposed obstacle detection model with the TEB local path planning algorithm.When an obstacle category is detected,the obstacle category information is sent to the local path planner,which sets different safety distances for dynamic and static obstacles and re-plans the local path accordingly.Experimental results show that the proposed improved local path planning algorithm can improve the safety of the inspection robot during movement.The experimental results indicate that the improved YOLOv5 model achieves an average precision(m AP)of 91.18%.Compared with the original YOLOv5 object detection model,although the detection accuracy has decreased by 1.50%,the model size has decreased by 23.20%,and the detection speed has increased by 12.6 frames per second.This demonstrates that the improved obstacle detection model is more suitable for real-time detection of obstacles in substations.The improved D* algorithm has shown some improvement in path smoothness,with an average path reduction of 3.69%.The fusion of obstacle detection model and TEB algorithm in the local path planning algorithm enables the inspection robot to maintain a safe distance from obstacles of different categories,improving the safety of the inspection robot during its movement process.