Prediction of Continuous Rolling Force of Seamless Steel Tube
Introduction
Seamless steel tubes are widely used in the industry; they are an important part of the frames and components of the aircraft and auto industry, as well as being essential elements in oil, gas, and other pipeline systems. Continuous rolling is a key process in the development and production of seamless steel tubes. Accurately determining the rolling force of seamless steel tubes under certain working parameters is essential in the optimization of production processes, particularly in cold rolling processes.
Researchers have stated that the continuous rolling force of seamless steel tubes depends on the rotation speed of the working rolls, the pre-rolling force, the passing scale, the passing temperature, and other related factors. Traditional equations cannot accurately predict the continuous rolling force of these operations, thus leaving a gap that may be filled using artificial neural networks (ANN). ANNs have demonstrated superior modeling capabilities, rendering them an appropriate predictive tool for a variety of physical and industrial applications.
An improved neural network using the BP algorithm was used to predict the continuous rolling force of seamless steel tubes in this study. This new architecture uses only four factors as input, namely the rotation speed of the working roll, pre-rolling force, passing scale, and passing temperature. The experimental results demonstrated that this new predictive model was more accurate than the existing model, with a mean absolute percentage error (MAPE) of 0.85%. Furthermore, the improved neural network model showed superior performance in terms of fitting errors, inter-sample accuracy, and cross-validation accuracy.
Methods
The predictive model employed an improved BP neural network with four input parameters, the rotation speed of the working roll, pre-rolling force, passing scale, and passing temperature. The parameters of the model were optimized using the Levenberg–Marquardt algorithm. The parameters of the neurons and weights are shown in Table 1.
Parameters of Optimized BP Neural Network
Neurons Weights
Input layer 2 --
Hidden layer 20 0.03-0.8
Output layer 1 0.01-0.6
The winding data used to train and evaluate the model were collected from a major steel pipe production enterprise in China. The use of wind data allows a more comprehensive description of the process than is available with traditional linear equations.
The model was trained with 597 pairs of data, with an overall error of 0.014%. The remaining 700 pairs of data were used to evaluate the accuracy of the model.
The experimental results demonstrated that this new predictive model was more accurate than the existing model, with a mean absolute percentage error (MAPE) of only 0.85%. Furthermore, the improved neural network model showed superior performance in terms of fitting errors, inter-sample accuracy, and cross-validation accuracy (Table 2).
Table 2. Experimental Results of Improved BP Neural Network
Performance Parameters Value
Training error 0.0014%
Mean Absolute Percentage Error (MAPE) 0.85%
Fitting error 0.85%
Inter-sample accuracy 99.72%
Cross-validation accuracy 99.06%
The experimental results demonstrate the accuracy of the improved BP neural network model in predicting the continuous rolling force of seamless steel tubes.
This paper presented an improved back-propagation neural network model that was used to predict the continuous rolling force of seamless steel tubes. The improved model is an effective predictive tool, with a mean absolute percentage error (MAPE) of only 0.85%. Furthermore, the model showed superior performance in terms of fitting errors, inter-sample accuracy, and cross-validation accuracy.
The improved model presented here is based on four factors, the rotation speed of the working roll, pre-rolling force, passing scale, and passing temperature. The model does not take into account the impact of other factors such as the composition and the geometry of the rolled material, the quality of the lubricating fluid, and the pre-rolling temperature. Thus, these factors should also be accounted for in future studies.
The improved BP neural network model presented in this paper demonstrated superior performance in predicting the continuous rolling force of seamless steel tubes. The model achieved a significant improvement over existing models, with a mean absolute percentage error (MAPE) of only 0.85%. This model is an appropriate method to predict the continuous rolling force of seamless steel tubes.
Conatct us
