Research On Intelligent Control Of Excavators Based On Neural Network

With the promotion and application of neural network technology in various fields,the use of neural network technology to improve the level of intelligent excavators has become an important research content in the field of mechanical control.That can add new vitality into the development of this traditional construction machinery.According to the requirements of different task phases,starting from the whole process of the excavator's autonomous work,,the intelligent control algorithm of the excavator is designed using the LSTM and DDPG neural network in this paper.We attempt to integrate the algorithm of the autonomous navigation and trajectory control flow of the excavator and explore the feasibility of the application of the neural network in the intelligent excavators.First of all,aiming at the requirement that the excavator needs to obtain the target from the image and cannot lose the target due to relative motion,a visual tracking system based on the LSTM network is designed.The supervised learning method is used to learn the habit of human observation of moving objects.Compared with traditional visual tracking algorithms,visual tracking algorithm based on LSTM network can get rid of the limitations of certain types of tracking objects and achieve stable tracking of targets in motion.It can not only provide control parameters for autonomous navigation,but also improve the intelligence degree of the way of input target to excavators.The secondly,an autonomous navigation algorithm based on DDPG neural network is designed for the excavator's task of autonomously navigating to the work site based on limited environmental information and the need to control the continuous physical quantum such as speed and direction.By making use of its characteristics of trial and error correction,DDPG neural network makes mobile path planning according to its own sensor data and the output of visual tracking system.Compared with the traditional autonomous navigation algorithm that divides the environmental information and continuous control amount into discrete physical value.The autonomous navigation algorithm based on the DDPG neural network can better understand the environment and meet the requirements of precise control,comply with the intelligent and accurate future development trend of excavator autonomous navigation.Finally,in view of the trajectory control problem in the movable space of the excavating device,a trajectory task control algorithm based on DDPG neural network is designed to continuously control the angles of the boom,stick,and bucket to realize the bucket-to-arbitrary operation.Carry out a track-fitting experiment on the tooth arbitrary trajectory.Experimental results show that the excavating device can realize self-adaptive tracking of arbitrary trajectories in the motion space.The algorithm is derived from the neural network of autonomous navigation algorithm and the excavating of simulation environment and parameter configuration files.Compared with traditional excavating trajectory control algorithms aiming at obtaining mechanical benefits such as maximum digging force,excavating trajectory control algorithm based on DDPG neural network aims at improving the efficiency of completing the task of excavation trajectory,making the excavation trajectory control algorithm understand humans' instruction more intelligently.The excavator control study of this paper is based on the excavator operator's construction method,and divides the excavate task into three subtasks: visual target input,autonomous navigation,and excavation trajectory control.The LSTM network is used creatively for visual tracking and provides control parameters for subsequent autonomous navigation.In addition,the continuous-valued DDPG neural network is used for excavator autonomous navigation and excavation trajectory control which has breaking traditional control algorithms' limit that can only solve one type of control problem.It has finished initial integration of multiple control algorithms.