Tension soft sensor of continuous annealing lines using cascade frequency domain observer with combined PCA and neural networks error compensation

In continuous annealing processes, strip tension is an important factor which indicates whether the annealing line works steadily. The strip tension detection in the continuous annealing process, therefore, is essential for the reliable and stable operation of the units, and will help to improve the quality of strip products. However, in real annealing processes, only a limited number of strip tensions can be measured. Since installing tension sensors everywhere will be unrealistic and expensive, in this paper a cascade reduced-order observer method with principal component analysis (PCA) and neural networks (NN) compensation is proposed to estimate the unknown tensions between each two neighboring rolls so as to monitor the tension profile for the whole line. When the main observer model is established, the influences of strip inertia and roll eccentricity on tensions are taken into account for better estimation. Moreover, when the error compensation is designed, PCA scores are used as the inputs to the NN. The application results show the effectiveness of the proposed method, with the estimated tensions used for the analysis of the strip-breaks fault.