Research on slag layer identification in soft blowing of ladle based on image processing technology and numerical simulation

Abstract

The research integrates image processing techniques and numerical simulations to investigate the identification of slag layer in ladle operations. A three-dimensional multiphase flow model was employed to study the gas–liquid flow and slag layer dynamics in soft ladle blowing, validated with industrial trial data on slag bulge ratio. This data formed the basis for surge classification in image recognition. Industrial camera images of slag layers are processed through greyscale and optimisation to enhance the detection accuracy of target features. The VGG16 convolutional neural network model was utilised for deep learning of pre-processed images, enabling the identification of slag layer characteristics at different surge levels. The Adam adaptive optimisation algorithm was employed for training, aiming to improve recognition accuracy and processing speed. The experimental results demonstrated a recognition accuracy of 99.16% for the slag layer identification.