Study On Soft-sensing Model For Ash Measurement Of Floatation Tailings Based On Gray Features Of Image

As there is less, various, and finer coal produced in China and the market requires coal of higher quality, the floatation technique has become an important part of coal preparation. At the same time, the automatic floatation level is the bottleneck for development of the floatation technique. At present, floatation results are manually judged based on foam layers and colors of the floatation tailings to make improvements. With development of machine vision, a creative approach is developed for testing the floatation process. In this paper, the author studies on online detection of tailing ash with feature values of gray images taking the floatation technique of XUEHU Coal Preparation Plant as an example.In order to obtain high-quality samples, the author optimizes the image collection system and researches on running conditions of the system. Firstly, the color of the light source is changed and feature values of sample images are compared under different light sources. As the author discovers, it is under the white light source that the sample images are the most easily identified. As the light resource of the collection system are required to be on for a long time, be stable, and give out less heat, the white LED light is chosen. Secondly, comparing four ways about how to place the light source and camera, the author chooses to place two light sources in a symmetrical manner based on requirements for image quality and actualities of the image collection system. The two symmetrical light sources can provide well-distributed light, avoid large light spots, generate non-deformed images, and do not need much cost. Additionally, system performance is studied and it is found that light intensity decreases after the light source is turned on and about 50 minutes later the intensity becomes stable. In order to prevent samples from settling, a magnetic agitator is added to the system. After tests are carried out, it is found that when its speed is kept at 800r/min the samples are not settling and the abnormal pixel rates of the samples are less than 2.5%. In order to ensure that feature values of gray images are representative, the author optimizes the method for collecting the feature values. First, the area of the largest light spot in an image is used for judging whether the image is qualified. Images with the largest light spot of which area is bigger than 1000 pixels are deleted directly. These images are not used to collect feature values of gray images. Second, to remove abnormal pixels in an image, the median filtering method is adopted to preprocess the image so that it is ensured its feature is representative. At the same time, the author studies the impacts of physical features of floatation tailings on sample images. In the paper, the single factor test and orthogonal test are used to research changing trends of features of the sample images when different ash, concentration, and granularity are provided. The result of the single factor test indicates that when the ash of samples is changed, the feature values of the sample image have the same changing trends. Therefore, it is feasible to predict ash of floatation tailings based on the gray average of an image. Additionally, the density and granularity of samples have some impacts on the gray average of sample images. In order to quantify the impacts, average area and numbers of light spots are obtained as substitution variables of the density and granularity of the samples. Then variance analysis is conducted in the orthogonal test to conclude that the impacts of physical sample features on the gray average of an image are: ash > granularity > density. At last, the values of eight features, including gray average, variance, skewness, kurtosis, energy, entropy, average area of the light spots, and number of the light spots, are used as inputs of the soft measurement model.To ensure that the model is accurate and the calculation time is short, the PCA analysis is conducted for the eight feature values. The analysis results indicate that the first three feature dimensions contain more than 95% of the features. Therefore, the three feature dimensions are used as the inputs of the model for predicting floatation tailing ash. 214 groups of experiment data are got via sample selection to develop the SVMR and GA-SVMR models for predicting floatation tailing ash. Error analysis is undertaken indicating that the relative errors of the two models are within 15%. Furthermore, the prediction stability of the GA-SVMR model is better than the SVMR model.Finally, the author makes conceptual design for an online prediction platform of floatation tailing ash, mainly involving the sample collection device, PLC control solution, and automatic image collection, automatic image processing. All of these facilitate application of the online measurement device for floatation tailing ash in factory scenarios.In this paper, an image collection system is designed. In addition, the soft measurement method for floatation tailing ash is put forward. And good results about the system and method are obtained in the laboratory, which prove that the rationale is correct. The relative error that is predicted is within 15%. The soft measurement model can provided feedback signals for the automatic control system, which promotes development of the floatation technique.