Research On Vision Based Compensation Control Of Hot Continuous Rolling Process

The production of sheet and strip steel mainly relies on the hot strip rolling process,of which roughing is an important process.The rough rolling process often generates the problem of slab bending,which has a complex mechanism and contains a large number of nonlinear and coupled relationships,and there is no practical application of mature automatic control methods for asymmetric slab shape problems.At present,most of the production process relies on manual compensation of roll slit tilt according to the size and direction of slab bending,but due to the influence of manual experience and technical factors,it is impossible to give accurate compensation of roll slit tilt,resulting in insufficient control accuracy and easy to produce sheet quality problems,which can seriously affect the quality of finished strip steel products.In order to achieve high accuracy detection and prediction of slab bending and accurate compensation of roll seam tilt compensation value,a fusion method based on machine vision with machine learning and fuzzy model is studied.Firstly,to address the problem of inaccurate detection of the size of slab bending,a machine vision-based slab bending detection algorithm is established.Using machine vision method for slab bending measurement,using image processing technology,the pixels of the edge of the rolled part can be obtained to indirectly determine the size of the slab plane,determine the slab bending direction and calculate the slab bending size based on the centerline offset.Second,to address the problems of lagging slab bending detection and low prediction accuracy,a slab bending prediction model based on machine learning algorithm is established.The random forest algorithm and support vector machine in machine learning algorithm are used to establish the slab sickle bend prediction model respectively.The selection of the parameters of the machine learning algorithm has a great influence on the accuracy of the prediction model,and it takes some experience and time to choose the optimal parameters by repeating the experiment continuously.The experimental results show that the prediction model can effectively predict the subsequent slab bending amount,and the random forest prediction model based on differential evolution optimization achieves96.3% accuracy in the range of ±3mm.The support vector machine based on Bayesian optimization achieves an accuracy of 99.25% in the range of ±2 mm,and all the above proposed machine learning models can meet the field requirements.Finally,the compensation model based on fuzzy model is established for the problem that the manually given roll slit tilt compensation value is empirical and uncertain.The fuzzy model is established by combining expert experience and historical data,with the slab bending and thickness values as input variables and the roll seam tilt compensation value as output variable.The experimental results show that the fuzzy model can accurately give the compensation value of roll slit tilt.In order to solve the slab bending problem in advance,the prediction model and the measured bending amount and thickness value are proposed as the input of the fuzzy model,and the two roll seam tilt compensation values are obtained as the roll seam tilt compensation values of the current pass.The simulation experiments show that the method can effectively reduce the slab bending.