| The steel industry has always been an important material foundation for China’s development of the national economy and national defense construction.Industries such as construction,bridges,and automobile manufacturing have increasingly demanding requirements for hot-rolled strip products.Yield strength,tensile strength and elongation are important indicators reflecting the mechanical properties of steel.However,due to the dynamic,complex,and non-linear characteristics of the steel smelting process,the performance prediction of hot-rolled strip has always been a difficult problem in metallurgical automation.The prediction of mechanical properties of traditional steel is carried out through destructive tests after random sampling.But this method consumes a lot of time and energy.Deep learning is a new research direction in the field of machine learning.In recent years,it has shined in various fields.Deep learning is favored by more and more scholars for its excellent feature extraction capabilities.The most representative models in deep learning are convolutional neural networks and recurrent neural networks.After random initialization,the convolutional neural network can automatically learn the features of the data from massive data.Its local connection mode and weight sharing characteristics effectively reduce the number of parameters that need to be calculated in the model and reduce the complexity of the model.Recurrent neural networks can learn the relationship of time series from data,and can realize the function that traditional neural networks cannot remember.Based on the above analysis,this thesis proposes to apply deep learning technology to the prediction of mechanical properties of hot-rolled strip.First,based on the convolutional neural network,a performance prediction model for hot-rolled strip is established.The yield strength,tensile strength,and elongation are taken as research objects,and the prediction model from the chemical composition of steel and hot rolling technology to mechanical properties is realized.A large number of comparative experiments were conducted on the structure and parameter selection of the convolutional neural network model to optimize the structure of the model,which verified the effectiveness of the convolutional neural network model in predicting the mechanical properties of hot rolled strip.Next,considering that the traditional neural network model cannot reflect the gradual and continuous nature of the hot rolling production line,a method for predicting steel properties using a recurrent neural network is further proposed.Finally,through comparison experiments,it is compared with the Support Vector Machine and BP neural network models.The experimental results show that the algorithm proposed in this paper can effectively predict the mechanical properties of steel,and is superior to traditional models in prediction accuracy and generalization ability. |
