Research On Visual Control Methods For Cleaning Robot Of Large Condenser

Large condenser is one of the key heater exchange equipments in heat-engine plant and petrochemical industry. It plays the role of cold source in the thermodynamic cycle for large steam turbine which can decrease the exhausts pressure and temperature of the steam turbine in order to enhance the cycle thermal efficiency of the turbo generator. For the unclear cooling water and chemical reactions during heat exchange when the condenser is running, the fouling which is not favorable for heat transfer is accumulated in the inner wall of condenser tube. These fouling have produced such harm:severely decreases the heat exchange capacity of the condenser; lowers the vacuum degree; increases energy consumption and probably leads to accidence due to the blockage of condenser tube and followed by erosion and perforation.First of all, the paper explains the necessity of fouling monitoring and fouling cleaning by investigating the cause of fouling accumulation and the influence on heat transfer performance of condenser. Then it analyzes the advantage and disadvantage of the ordinary fouling measurement and cleaning methods. Afterward the concept of the cleaning robot is proposed and introduces the principle of cleaning robot for large condenser based on a mobile manipulator. In addition we describe the research methods and research content of visual control.When fouling factor is greater than a set value, cleaning robot would automatically move to the area waiting to be cleaned and using the center coordinates of tube to implement high-pressure water jet cleaning, so condenser tube edge detection is a key step to achieve precise positioning for condenser cleaning. We make analysis of the existing edge detection algorithm and depend on the edge detector three criteria proposed by Canny, make a detailed analysis for two-dimensional odd Gabor wavelet applied in step edge extraction and done a careful comparison with the existing edge detection algorithm. As a result, two-dimensional odd Gabor wavelet is proved better than edge detection algorithm when used in condenser nozzle edge detection. In addition, the complexity of the underwater light is taken into account, the images edge may miss or incomplete after the edge process; and then we studied the edge fitting algorithm based on least-squares method, finally proposed improved random Hough transform algorithm to get the condenser tube center coordinates.If cleaning robot is wanted to autonomic move in the condenser water tank, first of all, the key problem is to solve the intelligent navigation. As the general camera visual angle is only 40-50 degrees which can not meet the need of the Omni-directional requirements, the paper proposed Omni-directional camera visual navigation based on hyperbolic Catadioptric and deduced in detail the camera catoptrics'principle, epipolar geometry, image unfold model, then Kruppa equations is applied for camera parameters calibration. Robot navigation is obtain the target image by fixed and mobile Omni-directional cameras and calculate epipolar geometry by means of feature points matching which could obtain the error between current location point and ideal location point according to change in the imaging plane. And then we can calculate the rotation angle and translation distance which mobile robots need to carry out. Therefore analysis the kinematics and dynamics model for the underwater robot, finally, the cleaning robot left wheel and right wheel controlled by back-stepping control algorithm, simulation experiments show that robot can move to the specified location accurately and quickly.when the creeper truck arrive the designated location, if directly use the nozzle coordinate for mobile manipulator would lead to location deviation because the water hydromechanics and mechanics process precision, thus affecting the cleaning effect. In order to solve this problem, this dissertation proposed manipulator visual servo control algorithm based on epipolar geometry, which extracts the feature points from the nozzle edge by mean of Gabor envelope algorithm. After calculation the fundamental matrix and the error between feature points and epipolar line, the task function is applied to split the control variable into two decoupled tasks. The primary task is make feature points converge on the epipolar lines using the rotations only, the second task allows feature point to slide along the epipolar lines corresponding to the desired position, so the Cartesian trajectory of the robot's manipulator is perfectly predictable, first of all the robot manipulator begin to rotate, when obtains the same orientations with the expectations location, then translate to the expect location. The new methods does not depend on anymore prior knowledge of the 3D structure scene, it only need simple control law and could get wider working space than eye-in-hand system. Simulation result on robotics toolbox for matlab and epipolar geometry toolbox shows that the method can obtain good control effect. All feature points finally convergence to the desired position and the total error converges to zero within a finite time. For the visual servo system, the visual performance of the controller depends on the system extracted image features, in order to avoid the problem of local image features beyond the visual scene. this paper using global image features-image moments as image feature information, avoiding the complex features match, Then deduced the jacobian matrix base on image moments. the visual control is realized with unknown depth of object, the camera focal length and does not require precise calibration of external parameters of the camera case. The simulation results show the image moments can used as image-based Visual servo control system.In the end, according to the specific requirements of cleaning robot, we proposed locally weighted partial least squares regression algorithm for prediction of condenser fouling, which fitting new measurement data by means of partial least squares regression analysis in old measured dada with a number of local models, then apply adaptive algorithm to adjust optimization model parameters and weighted coefficients of multi-model. The algorithm is a very good solution to the mutual impact of the new and old measurement data in order to adapt the condenser water quality changes and the device parameters dynamic changes. Then introduce in detail the cleaning robot motion mechanism, rotation mechanism, the chassis mainframe mechanism, electric putter, robotics manipulator, high-pressure nozzles and other mechanical component, and then introduce in detail the cleaning robot control system hardware composed by industrial computer, touch screen and PLC, as well as control system software composed by configuration software development, fault alarm systems, high-pressure water system, also provide specific model and parameters of some key instrumentation to ensure that cleaning robot fulfills cleaning task with high efficiency and stability。Finally, the main innovations of the thesis are summarized and the fields for further research are expected.