I'm sure no one will be surprised that I am still working on this. I'm still not satisfied, and it is a very challenging problem.
This week I made a couple of "breakthroughs". First, the Mastech LX1010B luxmeter I have has a very easy to modify sensor, because the photodiode is enclosed in a cylinder of plastic almost exactly 1.25 inches in diameter, so it will fit well inside a 1.5 inch diameter acrylic tube. Also, the assembly is very easy to take apart and put back together. Plus, the cosine filter "dome" is just a bit bigger than 1.5 inches in diameter, and isn't permanently attached to the housing - it pops out. So, the photodiode and cosine filter are easy to fit into a short piece of 1.5 inch acrylic tube, which is easy to buy in one foot lengths on no ebay links allowed
Next, after a week of trying to find a combination of Roscolux filters that will convert the lux readings to PAR readings it finally occurred to me that I don't need to do that. I can just convert it to read 10XPAR instead, which takes far fewer filters, making it much easier to find a set that do exactly what I want. And, when a sensor good for reading 20,000 in its indicator is going to be marginal for accurately reading 20 on that indicator. By shooting for 10XPAR, it will be reading 200, or 10% of the lowest scale, and by putting the scale selector on the 20,000 range, it reads PAR with no conversion needed.
Today I found a workable method for selecting the filters, and got it down to only 4 filters. Those make it read exactly the same as the Quantum PAR meter, and the spectral response is also good:
By requiring only 4 different filters, the Rosco sample book goes a lot farther for modifying these. I can get at least 4 sets of filters from one book, or buy just 4 sheets of filters, and have enough for 100+ PAR meters.
By the middle of next week I will have my 1.5 inch tubing, and be ready to work out the last details for converting it to a true, waterproof PAR meter. One absolute rule for this is to do nothing to interfere with the optical path in the sensor, besides adding the filters.
I have also finally figured out that my connectors that allow me to switch sensors aren't a workable idea. When working with such low voltages and currents, the contact resistance in the connectors is too big a problem and a variable one at that. Also, there is no reason to use more than one sensor, if I do this well.
My biggest problem will be finding another equally interesting project when I finish this.