Super-resolution techniques for stationary and nonstationary signals in machine direction analysis

Abstract

Many flat-sheet-form products such as paper, plastic and metal sheets, steel strips, and thin-film electronics are produced with roll-to-roll (R2R) processing, where cylindrical rolls shape the final product. The rolls heavily influence the quality of the product surface, and problematic rolls can leave undesirable periodic imprints on the product according to their angular frequencies, so detecting them is an important task. This can be done using signal processing techniques, particularly frequency analysis, on measurement signals along the product (machine direction). In the paper machine and steel rolling mill, many rolls have close frequencies, so there is a need for super-resolution analysis techniques. These techniques can be applied directly to stationary signals such as paper measurement signals in the paper machine, where the rolls operate at fixed parameters; but they cannot be applied directly to nonstationary signals where the frequencies evolve over time, such as in steel rolling due to the automatic gauge control system.

This thesis reviews and compares a few high-resolution frequency analysis techniques for detecting faulty rolls, as well as modifies the techniques for nonstationary steel rolling signals. A new approach, coined 'stationarization', is introduced which allows many techniques to analyze nonstationary steel rolling signals. Simulation experiments were conducted to compare and rank the techniques, before they are applied on real measurement signals. Compressive Sensing, one of the super-resolution techniques, is found to have improved resolution and accuracy compared to techniques used in previous research. The final result of the thesis is suggestions for the choice of techniques based on different scenarios and usages.