Premium
Statistical procedure to identify sources of extrusion line variation
Author(s) -
Stevenson J. F.
Publication year - 1989
Publication title -
polymer engineering and science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.503
H-Index - 111
eISSN - 1548-2634
pISSN - 0032-3888
DOI - 10.1002/pen.760292213
Subject(s) - plastics extrusion , extrusion , offset (computer science) , term (time) , variation (astronomy) , statistics , design of experiments , biological system , mathematics , materials science , computer science , physics , quantum mechanics , astrophysics , metallurgy , composite material , biology , programming language
A statistical procedure is proposed to identify sources of variation in the operation of complex extrusion lines. This approach involves separating multichannel extrusion line data, as necessary, into appropriate time scales; identifying significant cross correlations among various pairs of channels; and then comparing the pattern of the observed cross correlations with those expected for possible sources of variation. This procedure is able to detect unsuspected relationships among processing variables and to provide a rational basis for upgrading extrusion line performance. The pattern generation and comparison methodology could also be useful for expert systems. To illustrate the procedure, 11 channels of data from a dual extrusion line with metered feed were collected and analyzed. Since two sources of variation on different time scales were evident, the data were divided into long‐term and short‐term components using statistical methods. Cross correlation functions among pairs of variables were used to estimate offset times and identify variables that were significantly correlated. To be considered significant, the extremum in the cross correlation function had to occur at a realistic time offset and have an absolute value exceeding a prescribed threshold. The observed cross correlations between pairs of short‐term and of long‐term components were compared with the expected correlation patterns for typical process disturbances. By this procedure, the cause of the short‐term trends was identified as surging primarily in the larger extruder; a particular long‐term trend was attributed to a property change in the feed material to the smaller extruder. This analysis formed the basis for a novel process control device to minimize the consequences of short‐term variation in extruder output.