Rolling force prediction in cold rolling process based on combined method of T-S fuzzy neural network and analytical model

In the cold rolling process, inaccurate rolling force settings and the resulting strip thickness fluctuations and other quality problems occur, reducing the yield and product quality. To improve the accuracy of rolling force prediction, this paper proposes three methods to combine a T-S fuzzy neural network and rolling force analytical model based on their advantages and characteristics, to construct a combined rolling force prediction model, and to fully utilize the features and benefits of each model for rolling force prediction. The model's performance is evaluated by selecting the mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean squared error (RMSE). The model experiments with historical production data obtained from industrial sites. The experimental results show that the combined prediction models have a more robust rolling force prediction capability than the T-S fuzzy neural network model alone, especially the combined form of using the calculated value of the rolling force analytical model as the input to the T-S fuzzy neural network without destroying the self-learning of the rolling force analytical model, which has better calculation accuracy and reliability for rolling force than other models. The model can provide an essential reference for the online prediction of cold rolling force and high precision rolling production and has high usability.

Automated summary

This paper proposes three methods to improve the accuracy of rolling force prediction by combining a T-S fuzzy neural network and a rolling force analytical model. The experimental results show that the combined model has a stronger prediction capability for rolling force, especially when using the calculated value from the rolling force analytical model as input. The model provides an important reference and has high usability.