Estimation of finish cooling temperature by artificial neural networks of backpropagation during accelerated control cooling process
| Autor: | Yong Tae Kang, Hwan Suk Lim |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Fluid Flow and Transfer Processes
Air cooling Artificial neural network Mechanical Engineering 02 engineering and technology Sigmoid function 021001 nanoscience & nanotechnology Condensed Matter Physics Standard deviation Backpropagation 020501 mining & metallurgy Nonlinear system 0205 materials engineering Control theory Water cooling Artificial neuron 0210 nano-technology |
| Zdroj: | International Journal of Heat and Mass Transfer. 126:579-588 |
| ISSN: | 0017-9310 |
| Popis: | Artificial Neuron Networks (ANN) is considered one of the most practical technologies in the fields of intelligent manufacturing. In this study, the conventional heat transfer model and multilayer ANN analysis are compared to analyze the accelerated control cooling process, and the accuracy improvement of finish cooling temperature prediction by the ANN is evaluated. The temperature prediction error from the heat transfer model tends to increase with increasing the start cooling temperature in Curie temperature. It is found that the specific heat for low carbon steel shows a nonlinear tendency in Curie temperature. The ANN of backpropagation is applied to solve the nonlinear tendency of the specific heat. In the ANN analysis, the key parameters such as dimensions of plate, chemistry, start cooling temperature, air cooling time, water cooling time are selected as the input values. The hyperbolic tangent, sigmoid and linear functions are applied for the activation functions. The weights training was conducted 100,000 times, the weights were trained to satisfy the standard deviation of finish cooling temperature within 10.56 K. It was found that the accuracy from the ANN analysis was improved 2.74 times than the heat transfer model with least square method. It was concluded that the ANN with multilayer type could train the weights by the effect of the nonlinear trend of specific heat according to temperature. It is recommended that the heat transfer model should be replaced by the neural networks method of 3 layers (one input-layer, one hidden-layer, one output-layer) with the trained weights for the precise control cooling. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect