Research On Intelligent Monitoring Of Materials In Bucket Elevator Of Jigger Based On Binocular Machine Vision

The bucket elevator of jigger is responsible for lifting and transporting the refined coal,medium coal and gangue products washed by the jigger.The material quantity of bucket elevator and the coal content in gangue are both important monitoring indicators for the jigging separation process.When the former is too large,it can easily cause the bucket elevator to be pressed,resulting in deformation and damage of the equipment.The latter can reflect the level of jigging operation,causing waste of resources when it is too high.At this stage,these two indicators still need to be manually monitored.It is not only costly but also inefficient,which hinders the development of intelligent jigging coal preparation.Based on the actual needs of the coal preparation plant,an intelligent monitoring method of jigger’s bucket elevator material based on binocular machine vision was proposed.Combined with the instance segmentation model and binocular stereo matching algorithm,the three-dimensional reconstruction and volume prediction of coal and gangue in bucket elevator were completed.Besides,the coal content in gangue was predicted with empirical density.Firstly,the camera calibration experiment was carried out to obtain the internal and external parameters of the camera.The dataset of coal and gangue was obtained by image acquisition and registration.Based on the PSMNet network,a spatial global dual attention module and a multi-scale fusion module were introduced to construct an endto-end binocular stereo matching global fusion network(GFNet)for coal and gangue.Its performance on public datasets and coal and gangue datasets was superior to other networks such as PSMNet.After adding two modules,the endpoint error(EPE)index of GFNet was reduced from 0.819 to 0.785.On the KITTI2015 and KITTI2012 datasets,the disparity accuracy of GFNet in all regions was at least 0.17 % and 0.10 % higher than that of PSMNet.Secondly,the YOLACT instance segmentation algorithm was used to segment and identify the coal and gangue images.Then a model with good segmentation effect and high accuracy was obtained.In the validation set,the mAP @ 0.50 value of the rectangular box was 97.17,and the mAP @ 0.50 value of the mask was 87.46.The segmentation recognition effect could meet the requirements.Then,based on the results of the instance segmentation and recognition,the traditional SGBM method was used to obtain the depth map of the single-layer stacked coal and gangue.After that,the weighted least squares filtering algorithm was used to remove the burr noise.The three-dimensional reconstruction of coal and gangue was realized.The volume prediction model of the corresponding particles was established by using the thought of integral,and the minimum average error was only 2.37 %.At the same time,a volume prediction model was established by using GFNet.After comparison,the accuracy of the model was not significantly improved,which was0.18 %.In the multi-layer stacking experiment of coal and gangue materials in the bucket,the prediction accuracy and stability of the volume prediction model established by GFNet are better than those of the SGBM method.In the multi-layer stacking experiment of coal and gangue in the bucket elevator,the prediction accuracy and stability of the volume prediction model established by GFNet were better than those of the SGBM method.The average volume errors were reduced from 5.49 % and 6.50 %of SGBM method to 4.24 % and 4.19 %,respectively.It could be better applied to the volume prediction of multi-layer stacked coal and gangue.Finally,combined with empirical density,the coal and gangue volume parameters were used to predict the coal content in gangue.When there was a few materials in the bucket elevator,the average error of prediction was 9.19 %.When materials were stacked in multiple layers,due to the serious coverage of particles,it was not appropriate to calculate the index of coal content in gangue.According to the actual demand,the situation of coal content in gangue was divided into three states: normal,high reminder,and excessively high alarm,so as to monitor and guide the situation of coal content in gangue in jigging production.Based on the industrial field,the intelligent monitoring strategy of the materials in the jigger’s bucket elevator was formulated.A single machine comparative test was carried out.The average error of the binocular vision monitoring system in detecting the amount of materials carried by the bucket elevator was 10.55 %.It basically met the requirements of intelligent transformation of jigger in coal preparation plant.Besides,it had been industrialized and applied in the Zhuneng Heidaigou coal preparation plant.