I'd make the longest time a bit longer, so if you wish to have 19 min, make the longer one, perhaps 35 min. The idea is not to hit the exact target time with a test, but to generate enough data to the "right" and "left" of your target times that let you safely interpolate the actual, correct duration. If you feel that you are good with 18 min, and you wish to re-test, then you need enough data on both sides of the 18 min to be sure that, for example, 25 would not be better.

In general, choosing 5 dev times that follow the sqrt(2), ie, 1.4142 sequence helps to cover enough range for typical N-2 to N+2 needs. If your 18 min seems to be your N time, then I'd suggest a sequence of: 9, 12, 18, 25, 36 min. The key to the test is the interpolation of the correct times in-between the tested ones, based on the curves/contrasts that you would get from those tests, and those that you would desire to get. Ralph's method will lead you there, or you could follow many others. I've published curve-plotting software on this forum, too, which can also calculate the resulting times, based on your contrast requirements, but it can all be done, easily, by hand.