UBC Theses and Dissertations
Estimation of cross and machine direction variations using recursive wavelet filtering Aslani-Mahmoodi, Setareh
Paper machine data taken from scanning sensors are used to assess the quality of the two-dimensional sheet. Scanned data normally include significant noise components as well as aliasing effects resulting from the irregular sampling pattern of the scanning sensor. The separation of the cross machine (CD) variations from those in the machine direction (MD) is important for control purposes, but is compromised by interaction of scanning patterns with MD disturbance frequencies. Different filtering techniques have been proposed in the past. In this work some methods are presented to remove the noise, mitigate aliasing, and provide accurate CD profiles. The high frequency machine direction (MD) process variations present in the data, should not be allowed to distort the cross direction (CD) variations as far as possible, for the purpose of CD control. The filtering method presented here uses wavelet filtering for noise reduction, and takes advantage of a bootstrap recursive scheme for successively improving estimates of MD and CD variations in the scanned data, and separating MD and CD profiles. The approach also uses periodic sampling theory for interpolating the process variations, and recognizing the change in sampling pattern that occurs at different CD positions. Some sets of simulated data as well as several industrial data have been used to test the proposed algorithm. The advantage of using the simulated data is that in this case the true variations are known. The aliasing effect of the MD profile, especially for periodic MD signals of a period close to that of the scan rate, is another issue in filtering scanned data effectively. A method is presented to detect the aliased MD frequency, in the.filtered profiles. This test may provide guidelines for changing the scan time if necessary.
Item Citations and Data