Research On Control System Of Robotic Excavator Based On Machine Vision

Hydraulic excavator is a kind of engineering machinery which is widely used and has a complex structure. At present, many research institutes are working to improve the automation and intelligence level of excavator at home and abroad with the development of excavator technology. To realize the autonomously mining for excavator, the robotized excavator is the major orientation.In order to improve the work efficiency, reduce the labor intensity and adapt complicated work environment, how to realize the autonomously mining object will be the research focus to improve the level of intelligent of excavator.Work device and electric-hydraulic system model are established in this paper, researched the trajectory planning and simulation of work device based on the robotized excavator transformed from excavator. Designed a kind of auto-disturbance-rejection visual serving controller suitable for robot, combined with environment information getted by visual sensor, and validated the effectiveness of auto-disturbances-rejection visual servo through the simulation and experiment. Builded control architectures of excavator robot based on the behavior control, the given mining task wos decomposed to achieve autonomously mining. As follows are the several main research works:(1) Designed the transformation programme of electric-hydraulic system specified to PC02-1Komatsu hydraulic excavator, fulfilled the robotized and electric-hydraulic proportional technology modification for excavator. Installed the tilt sensors and pressure transmitter on the work device, installed the visual sensor on the machine frame. Robotic excavator’s xPC target control platform is set up in the environment of Matlab/Simulink and xPC target’s configuration method and work flow are presented.Used host PC-target PC’s external mode to realize robotic excavator’s real-time control, so can do dynamic models test conveniently,and acquire optimum control parameters real-time.(2) The work device of excavator robot similar to mechanical arm driven by hydraulic cylinder, used the kinematics theory of robotics to build the forward/inverse kinematics model of work device, and established the contacts with the position-pose space, joint space and oil cylinder space,to realize the control for robot in each space. Establishe the model of valve control cylinder system of excavator robot, and use RBF neural network method to identify electric-hydraulic system parameters. Use optimization strategy of RBF neural network,for inverse kinematic solution of work device,to realize trajectory planning of the end of bucket of work device.To improve the trajectory planning control precision of excavator robot’s work device while mining,established ANFIS inverse mapping model,select the input and output data between inverse mapping surface to train ANFIS structure, get corresponding joint angle according to the given expectation mining trajectory,and then be used to track expectation trajectory,the simulation result shows that tracking precision can well meet the practical requirements.(3) According to the characteristic of illumination condition influence to outdoor road image, presents decorrelating drawing transform method. In order to improve the image quality of bucket, use histogram homogenization method to enhance image.To remove the background influence of obstacle object image brought in image segmentation approach based on gratitude watershed transform to segment obstacle and object image from the excavator robot’s motion path image.In order to improve the identifying accuracy for bucket target,put forward the target images recognition method based on the moment invariants and improved BP network,so improved the recognition reliability of the bucket target.(4) Parameters calibration of camera is the basis establishing measurement system of machine vision, and the guarantee improveing vision’s measurement accuracy.Established the both inner and outer parameters imaging model of camera vision system of excavator robot,analyzed the nonlinear distortion parameters of camera, the calibration parameters for vision system of excavator robot are determined.Through sampling self-made checkerboard calibration template image,realized to extract template’s angular point based on OpenCV technology.And then calibrate the internal parameters matrix of the camera model,the radial distortion coefficient of nonlinear model of camera,the external rotation matrix and translation sector of camera,and the calibration parameter errors are given. Calibration results shows that error accuracy reach sub-pixel precision, can meet the demands of the vision system calibration of excavator robot and the vision measuring accuracy.The binocular vision system model of excavator robot is set up, and the model parameters are calibrated.Analysis the target image’s matching method of bucket of excavator robot, simulation research is performed on the binocular stereo visual match.(5) On the basis of stereo vision system’s calibration and image matching, researched the depth information measurement method of stereo vision, target image’s fixed position method moving object’s tracking method. Researched the object and it’s attitude recognizing method of bucket of excavator robot based on color marker tracking.Researched the image feature extraction and the location method of the bucket.(6) Researched the image jacobian matrix and image jacobian matrix estimation method based on improving neural network.In order to improve the autonomous mining capacity of excavator robot,the auto-disturbances-rejection visual controller based on image is designed to control the tip-position and posture,in x-z plane coordinate system, of the excavator robot’s mechanical arm which is composed of boom,stick and bucket. For the reason that has many parameters needed to be tuned for auto disturbance rejection controller, has mutual influence between the parameters and has the difficulty with parameters tuning, introduced particle swarm optimization algorithm to optimize the parameters of the controller.Because of the shortcoming that original particle swarm algorithm get into the local optimum late, uses niching particle swarm optimization algorithm to tune parameters of active disturbance rejection controller.The active disturbance rejection control system based on vision is established, and the simulated experiments were carried out.(7) In order to achieve autonomy excavation for excavator robot, builds the behavior control system architecture suited to excavator robot.Excavation behavior as a baseline, using the state-flow model, achieved state decomposition:excavation object, excavation task and excavation behavior stepwise.Collect object image, rake angle signal of arm and pressure signal of hydraulic cylinder, as the event or condition for trigger and transition between behavior states of state-flow.During excavation, according to the three kinds of mining environment which are sand material, surface block material and block material buried in sand, comprehensives visual position information, pressure and rake angle information, through fuzzy clustering discrimination, trigger mining action state flow model, process, distinguishingly and independently, the different circumstances encountered in mining, through controlling excavation action finally, achieve the autonomy excavation object.