What you've just described is a curvilinear relationship. This could be modelled in regression using a cubic function (i.e., pressure^3 or flow^3; I don't know which is your independent variable). However, be careful of overfitting; with enough polynomial terms, a regression line could be drawn to fit any data (for example, if you have 50 datapoints that are completely scattered around with no linear relationship/correlation, you could still draw a line using a 50th-order polynomial that goes up and down 49 times to fit all the data... however, it is very unlikely that that is meaningful) but should be constrained by your a priori theoretical and common-sense predictions about what the data pattern should be; if you had a theoretical reason to think the data would show this sort of relationship then it might be meaningful, but otherwise it might just be noise (indicative that there isn't a robust relationship between the two variables).