Previsão de temperatura de bobinamento de aços laminados a quente utilizando redes neurais artificiais

Abstract

During the production of steel coils in a hot rolling, the coiling material temperature control is of paramount importance to the process. Its variation can cause changes in the materials mechanical properties and microstructures, producing materials with nonconformities which may generate waste. This work aims to contribute to improving this process through the application of numerical modeling and computational intelligence in the estimate of the convective heat transfer coefficient from the run-out cooling table and predicting coiling temperatures. Firstly, the data of several coils were collected considering the run-out cooling table process variables and the achieved coiling temperatures. Then, using numerical methods and optimization, the convective heat transfer coefficient is determined for each collected sample. Finally, a neural network is applied to define the relationships between process variables (thickness, water ow, etc.) and the estimated convective heat transfercoefficient. The results are compared with other models found in the literature and they show that the proposed approach has superior performance. The estimated values can be used to predict coiling temperature and together with control techniques appliance to contribute to the material mechanical properties and microstructure improvements.