| Abstract |
From: Application of Artificial Neural Network Systems in the steel industry
During continuous casting of stainless steel so called breakouts may occur which are caused by disturbance of the solidification process within the mould. Breakouts may cause enormous costs due to the standstill of the plant and the costs by repairing the casting facility.
|
| Introduction |
The application of automatic alarm systems is intended to release the operator from the necessary direct surveillance of the measured signals at the continuous caster on the one hand, and has to provide a reliable and early warning for critical process states on the other. Tha alarm system has to make autonomous decisions based only on the available data.
Neural Networks provide a convenient way to implement alarm systems that are able to learn a specific task by examples, and that can easily be adapted to variations in the process.
|
| First steps |
Using neural networks is not always worth it, it has to be considered. Let's decide in the first steps.
|
| Objectives |
The objectives for the Neural Breakout Prediction System can be summarized in the following points:
- Reliable detection of stickers during the continuous casting process.
- Avoidance of the misclassification of conventionnal alarm systems.
- Avoidance of misclassification that are specific to the neural net.
- Invariance with respect to variations in steel quality.
- Robustness according to variations of the casting facility.
- Robustness according to bath-level and steel temperature variations.
|
| Application to Breakout Detection in Continuous Casting |
Theory is not everything, here is the application to our problem.
|
| Existng projects |
SENECA is an existing project that tends to solve the SDB problem.
|
| NeuroShell 2 tests |
Here is the first Visual Basic program I've made that uses a NeuroShell's network: XOR.
|
| Reports and Reviews |
Here are the reports I wrote for the University.
The first results I obtained.
|
