Research On Multi-robot Dispatching Method For Train Maintenance

With the rapid development of our country’s railway construction and the increase in the number of vehicles operating on the railway,the operation safety of high-speed trains has become the focus of attention.The inspection and maintenance of EMUs is one of the important links to ensure the safe and stable operation of EMUs.With the increase in the number of vehicles for inspection and maintenance,the workload and pressure of the workers also increase.The traditional "personal inspection and repair" method can no longer meet the maintenance requirements,and it is imminent to speed up the change from "personal inspection and repair" to "machine inspection and repair".With the rapid promotion of robot applications and the rapid development of artificial intelligence technology,intelligent train maintenance robots are gradually entering the field of train maintenance.The robot for overhauling trains has high flexibility and adaptability.Its application can improve the efficiency of maintenance,help the transformation and upgrading of railway maintenance,and promote the safe construction of railways.With the increase in train inspection tasks,a single robot can no longer complete so many tasks.The use of multiple robots at the same time requires unified planning and scheduling of multi-robot operations to improve the efficiency and utilization of robots.In this paper,the two core issues in the scheduling of multi-train inspection robots are studied separately: task allocation and path planning.First of all,based on three different optimization goals: the shortest total running distance of the multi-train inspection robot,the shortest time for the multi-train inspection robot to complete all tasks,and the comprehensive optimization of the total distance and total time.Considering the power constraints,the mathematical models of phase task allocation are established respectively.Because the traditional genetic algorithm has problems such as slow convergence and poor optimal solution in the train inspection scene,this paper proposes a "semi-selection and semi-crossover" population generation method,designing an adaptive selection operator to speed up population convergence;introducing the idea of quadratic adaptation in crossover mutation probability,and proposing a second mutation strategy to promote the generation of optimal individuals;an evolutionary compensation strategy is proposed based on the situation that the optimal fitness of the population remains unchanged for multiple successive generations to avoid the algorithm from falling into the local optimum.Secondly,according to the actual train inspection three-dimensional scene model,the sparse division method is proposed on the basis of the grid uniform division method to establish the path planning grid map,and then based on that in the actual application of the traditional A* algorithm in the train inspection scene,there are problems such as not considering the volume of the robot,the planned path does not match the direction of the robot’s movement,and the inability to avoid obstacles.Propose a virtual decision strategy to solve the problem of collision with local obstacles due to the lack of consideration of the robot volume when the traditional A* algorithm is used for path planning;Introduce the direction-changing penalty function to improve the evaluation function,solve the problem of the inconsistency of the A* planning path with the robot’s movement direction;adopt secondary path planning method and design dynamic obstacle avoidance strategy to solve the problem that robots cannot dynamically avoid obstacles.Finally,the algorithm verification of the task allocation and path planning of the train inspection robot is carried out by using relevant experimental data.Algorithm verification results show that: Compared with traditional genetic algorithm and adaptive genetic algorithm,the improved genetic algorithm in this paper has significantly improved solution efficiency and convergence speed,and it can allocate tasks reasonably according to different robot capabilities and states when assigning tasks;The improved A* algorithm considers the volume factor of the robot to avoid collisions with obstacles,and at the same time,it can dynamically perform the secondary path planning when encountering obstacles,avoid obstacles,and reach the destination smoothly.