Study On The Opening Degree And Control System Of Gate Group In Coal Washing Process Based On Machine Vision

There are many gates in the coal washery,some of which used for adjusting feed,discharge and water have not been fully incorporated into the centralized monitoring system due to the detection of opening degree.Currently,workers monitor the switching status and opening degree of these gates and adjust these parameters based on experience and knowledge,which is efficiency-low and time-consuming.In this paper,the machine vision algorithm is used to detect the opening of the gate,and the gate group control system is designed to realize the state monitoring and control of the gate group.First,the commonly-used image processing algorithms are introduced to locate the gate and detect its opening degree.On one hand,the template matching method is used for gate location.The experimental results show that it is not robust to light changes and therefore the performance of it is satisfactory only in the case of strict brightness control.On the other hand,the threshold method and the frame difference method are employed for opening degree detection.However,the threshold method is seriously affected by ambient light and yields poor detection performance,and the frame difference method is only applicable to the moving target and does not work in the case of stationary gate.Second,inspired by the successful usage of deep learning in many computer vision tasks,deep learning is employed for gate location and opening detection in this paper.The classical YOLOv3 is performed on the hand-made target detection dataset of gates and is evaluated in a range of experimental settings including various gate types,multiple gates,different brightness level,and with the presence of moving objects.Compared with the template matching method,the YOLOv3 is more robust and can locate all types of gates accurately.To detect opening degree,six semantic segmentation algorithms including FCN,PSPNet,Deeplabv3,CGNet,ICNet,and LEDNet,are performed on the hand-made semantic segmentation dataset of coal blending gates.To further measure the generalization performance of LEDNet,the effect of different brightness levels and interference on the detection performance are investigated.Besides,the detection results of LEDNet is compared with that of frame difference method.The experimental results show that deep learning semantic segmentation is suitable for gate opening detection with the support of sufficient samples.And the LEDNet semantic segmentation algorithm has good robustness and real-time performance for coal blending gate opening detection.In addition,a method based on the auxiliary laser line is proposed to detect the opening degree of the gate.Considering strong anti-interference ability of laser sensor,laser line is introduced in this paper to increase the robustness of image features,which is a low-cost machine vision engineering method.The emitter is installed on top of the gate and then laser line is projected onto the gate.The image processing algorithm is then employed to calculate the opening degree of the gate based on the detection of the laser line on the gate.Various laser line detection methods are chosen depending on the different characteristics of the pump gate and coal blending gate.For example,the pump gate uses the threshold method and the coal blending gate uses the HSV color to detect the laser line respectively.The experimental results show that laser line can accurately detect the opening of the pump gate whereas blurred laser line does not detect the opening of coal blending gate effectively.Finally,the monitoring system of the gate group in the coal washing process is studied.Based on the machine vision detection gate opening,the overall design of the gate group control system was designed,and the design of the upper computer opening monitoring software was completed.The thesis contains 83 figures,8 tables and 75 references.