Resarch On Detection System And Method Of The Ash Of Flotation Tailings

Coal flotation is the most efficient coal preparation of sorting and recoveryprocess for the coal less than0.5mm grain size. The good working of flotationcan directly enhance the coal quality, the recovery rate, and economic technicalindicators of the whole plant. Currently the site to determine the effect of theflotation is mainly based on experience of the operator, such as by observing thestate of the foam layer and tailings color to adjust the dosage, but the exist ofhuman subjective judgment differences and bias made it difficult to achieve andmaintain optimal flotation process. In this paper, the United States Wei Mukeflotation machine of Xiqu Preparation Plants is used for object, to study the online detection method of ash tailings based on machine vision as the studyobject.Firstly, the tailings slurry lifting device and image acquisition devices aredesigned according to the structure of the flotation tailings pond. About theLifting devices, bucket elevators is elected, and it is made up by the frame, chainplate, stainless steel chain, hoppers, dump festival, couplingsand speed motor.When lifting device tailings slurry is raised to a suitable location, the tailingsimage could be obtained through the image acquisition device. Image acquisition device consists of a light source, industrial cameras, and lightintensity sensor. System software can automatically capture the image and thelight intensity information tailings, and simultaneously record iris, exposuretime, and the correction parameter. Flotation coal, flotation tailings and gangueare used as a basis for the ash sample in order to study the relationship betweenflotation tailings and tailings slurry ash image features. According to it,63groups of ash from20%to60%, ash differential2%concentrations of thesample10gL-1,20gL-1and30gL-1is prepared. Through the test platform tocapture images of the sample and the light intensity information, each imagesample collection100and the corresponding intensity of incident light, reflectedlight intensity, a total of6300group of samples data. On this basis, theconventional feature of gray histogram obtained from slurry image is extracted.Then the mean gray level, variance, smoothness, skewness, energy and entropyof the six characteristic values is calculated, and added the incident and reflectedlight intensity, thus them in8-dimensional vector as the input of neural network,the corresponding gray value as output. But the test results show the tailings ashrecognition results are not satisfactory, the error is large, in the200test dataerror≥5ratio reached26%. In order to improve recognition accuracy tailingsash, the relationship between tailings sample images histogram and sample ashwas analyzed, found a significant correlation between the distribution statisticsof pixels in the gray zone and sample ash, rather than the mean gray. This end,take the gray value interval50-170, section interval is10, the number of pixels in12sub-section as a new input vector as the input of neural network, thecorresponding gray value as output for training and testing. The results showedthat the ash recognition accuracy has been greatly improved; all error≤5, inthe200test data error≤2ratio reached94%. In order to reduce the impact ofthe image acquisition unevenness we averaged10data which randomlyextracted from100dataas new100set of data, in order to eliminate the impactof random movement of tailings slurry to make the law more stable. In addition,the impact of tailings slurry concentration of ash recognition was studied, theneural network model was established at each concentration of cross-recognitionof each set of data, the results show that the identification of the ownconcentration data is better results, identify other concentrations is less effective.Two experiments is divided into two groups of20%ash and60%concentrationfrom5gL-1to30gL-1interval2.5gL-1, and found that high ash partition above20gL-1concentrations the mean gray impact on large, low ash zoneconcentrations less affected. Mixed three sets of data,the result of identifiedeach set of data is good, Concentration on the recognition accuracy is alsoreflected in them, High concentrations of high ash zone recognition effect is notvery good, low concentrations of low ash zone identify the general effect.Traditional training recognition feature extraction results are not satisfactory inthe same way. Overall, the absolute error identification using gray valuedistribution features in each concentration of ash within±5more than95%.This design system based on machine vision and the proposed flotation tailings ash recognition achieved good results in laboratory tests, prove the basicprinciple is correct, and high precision, also can provide feedback for automaticcontrol systems, create the conditions for on-line control system.