Closed bryanwweber closed 7 years ago
One possible fix for this is to limit the filter selection window to just the interesting part of the pressure trace. We should investigate if the noise frequency is a function of time.
It seems like a couple of things helped here for some sample data. First was using a first order Butterworth filter instead of second order. Another is trying to optimize the selection of the linear portion of the residuals curve by estimating R^2 of the linear fit and iterating the choices of the start and stop until R^2 > some value, say 0.95 or 0.99
Bad in this sense means that the agreement between the filtered pressure trace and the raw pressure trace is really bad for a short segment of time, usually near ignition. Usually increasing the filter frequency manually improves agreement. See below.