UBC Theses and Dissertations
Online detection of picketing and estimation of cross direction and machine direction variations using the discrete cosine transform Karimzadeh, Soroush
A scanning sensor at the dry end of a paper machine samples a noisy mix ture of both cross direction and machine direction variations. Although many techniques have been developed to separate these two variations and estimate a filtered profile, industrial practice has changed little over the years. In this work, a novel online Cross Direction (CD) /Machine Direction (MD) separation approach is developed. The method uses a fundamental prop erty of the CD variations to separate them from the MD variations. It is shown that the scanned CD variations build an even periodic function in the time domain and appear only as integer multiples of twice the scanning frequency in the frequency domain. Based on this property, the Discrete Cosine Transform (DCT) is utilized to separate the CD profile from noise and the MD variations. The performance of this method is verified through comprehensive simulation tests and is compared with existing wavelet and exponential filters. It is shown that the suggested method has better per formance in simulation scenarios. This is further validated by applying the method to industrial data. In the second part of this thesis, a novel picketing detection method is devel oped. The picketing pattern is associated with growing components in the spectrum of actuator profiles. A conventional change detection algorithm, CUSUM, is applied to detect these growing components. The method is verified in two simulation scenarios. It is shown that this method can detect and predict a picketing pattern before any picketing pattern is visible on the raw profile.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International