A conceptually straight forward exercise the technical difficulty of which is around mid-secondary school level maths.
Problem is that you MUST model the AFM lines, not the usual performance equations. The test of a good setup is that the delta between the accurately read line data and the computer output is both conservative and minimal. You cannot do a simple regression analysis as the errors will be too great. I did a lot of this stuff in the deep past - all good fun if excessively tedious.
Some thoughts -
(a) read off the points at a reasonable density consistent with the line shapes. Blow the AFM charts up. If you want to do it the hard way, rescale as photostats and then pore over the graphs with a magnifying glass reading off data points and setting up tables for the lines. Generally easier if you scan to high res and then do the reading in a vector package. Keep in mind the aim is to minimise the delta between the line and what you read while being conservative. Either way you end up with tabular data replacing the lines. If the AFM data was tabular to start with ... you can avoid this bit.
(b) set up a look up model. Either use the tabular data as a lookup table and interpolate or run regressions on sensible lengths of the lines. Don't even waste your time trying to determine regression coefficients for the usual squiggly lines .. won't work. Simple linear interpolation doesn't generally cut the mustard but there are standard techniques which help out
(c) test the model output against the read off tabular data. When you reckon you have an adequate balance between accuracy and conservatism move to the next activity
(d) set up the carpet models. That's just a computer analogy for reading the charts manually. Including enough relevant tests to capture errors and the like is where the time is spent.
(e) when you reckon you have it under control, run an every point comparison between the final model and the original read off data to check for errors.
Can you set this up in Excel ? Of course you can. Alternatively, use any of the appropriate programming languages.