System for recognizing gas flow distribution patterns in blast furnace centre based on computer vision

Abstract

Reasonable gas flow distribution plays a decisive role in the smooth operation of blast furnace smelting, but it is difficult to detect the gas flow distribution in blast furnace in real time. An intelligent prediction and identification system of central gas flow distribution based on infrared image of blast furnace and cross-beam temperature measurement is constructed(C-GFD). The system is mainly composed of two models, namely the image model and the prediction and recognition model. In the image model, three kinds of derived parameters, namely, central gas flow area, temperature and offset, are extracted by the image entropy and neighbourhood valley-emphasis (ENVE) Otsu, thermodynamic heat transfer and grey scale centroid algorithms, and then the statistical relationship between the change of image information and the distribution of gas flow is investigated. In the prediction and recognition model is established by the algorithms based on convolutional neural network long and short-term memory (CNN-LSTM) and Euclidean-weighted fuzzy C-mean clustering (E-FCM) to complete the prediction of the three types of derived parameters, and the prediction data is transferred to the recognition model to complete the recognition of the central gas flow distribution pattern. The test results show that the system provides real-time and reliable gas flow reference information for blast furnace operators with 95% accuracy in model prediction and more than 90% accuracy in pattern recognition of various types of central gas flow distribution.