Application of Neural Networks to Cold Rolling Continuous Proceses for fault detection and diagnosis

The continuous annealing process line is effective in producing high quality cold-rolled metal strips. The process line operates at high speeds. At such speeds, the tension profile of the strip needs to be controlled in order to prevent faults such as strip breaking or slippage. But, tension cannot be measured continuously along the line as only a limited number of sensors can be placed. So it is essential that the tension is estimated correctly to a sufficient degree of accuracy. The previous works focused only on modeling the system mathematically based on system parameters. But, they ignored the effects of non-linearities such as temperature of strip, gas flow etc. which are very difficult to model. This work takes up data-based approach to address those shortcomings and to come up with the estimation of the difference between the tension estimated from model-based approach and the actual tension, in order to reduce the error significantly. This hybrid approach has also been applied in real scenario and has been found effective. Index Terms Neural Networks, Principal Component Analysis, Data-based approach, Annealing Process