Mr Shorty
That was a good reality check for me. I am perhaps over complicating it.
The use of the logical check to bracket the cycle is a good start. I may
just have to play with the threshold.
(Just because I have a +ve number does not mean the refill has fully
finished.)
Or I could simply discount the first and last reading of the cycle.

I concentrated on using LINEST to deliver a rate of change because i
belived it would provide the best fit to the data.
In practical terms the difference between LINEST and and average of the dw
values is within the bounds of acceptable error.

The starting and finishing point method must be treated with caution
though. It is significantly influenced by the value of the particular
points used. The more scatter the bigger the risk.

Thankyou for taking the time to look at this and respond.
Cheers Rob