Floating State Division And Height Prediction Of Strip In Air Cushion Furnace With Hybrid Nozzles

The air cushion furnace with hybrid nozzles is the most typical structure currently,with large airflow impact force and high heat transfer efficiency.It is the key equipment in the heat treatment line of high-end metal strip.The floating height is the key factor affecting the production efficiency and product quality of the air cushion furnace.However,the floating height is difficult to measure because of the harsh environment in the air cushion furnace.In addition,there are stable state and vibration state in the floating process of the strip,which makes it more difficult to observe the floating height.Therefore,a new height prediction scheme is proposed in this article to divide the floating state of the strips in the air cushion furnace with hybrid nozzles and then realize the prediction of the floating height.First,considering the high cost of sample labeling and the limited number of labeled samples,a two-step time-series strip floating state division method based on empirical semi-supervised support vector machine is proposed.This method consists of two steps: rough division and fine division.To begin with,the rough division of strip floating state is realized by introducing the empirical knowledge of strip floating process into transductive support vector machine,which reduce the optimization time of hyperplane and make the classification model more reliable.Then,based on the rough division results,the samples that do not conform to the time-varying characteristics of actual process data in the rough division are divided twice by the sum of quadratic errors,which is called the fine division.Second,the time and variable information fusion based density clustering for strip floating state division is proposed.In this method,the dataset is constructed into covariance matrices,and instead of Euclidean distance,the distance between the covariance matrices is introduced into the calculation of the local density of the samples to capture the dynamic characteristics of the process.The time interval information is introduced into the calculation of the distance of the samples to avoid the false division caused by noise.Third,according to the process characteristics of stable state,a hybrid model combining mechanism model and error compensation model is proposed to predict the height of stable floating strip.The mechanism model is a height prediction model based on the ground effect theory and force balance equation under the working condition of stable state in the air cushion furnace with hybrid nozzles.The error compensation model is a data-driven model based on XGBoost to compensate the errors generated by the mechanism model.Finally,according to the floating characteristics of the vibrating strip,the XGBoost model is used to predict the maximum and minimum floating heights of strip.The experimental results on the existing air cushion furnace with hybrid nozzles show that the proposed height prediction scheme achieves good performance.