One of Professor Dworak's graduate students fixed Modelsim! The fix was actually very simple, instead of the command "run" to run the program it was neccessary to use "run -all". No one knew how to use Modelsim well enough to find that error so it took quite a lot of trouble shooting to fix. Once we got Modelsim working, we were finally able to analyze data. I ran randomized test sets in groups of 10, 50, 100, 1000, and 5000. Only when we ran the 8 test sets each 5000 times did we get complete coverage. Obviously these results are terrible as one cannot load 40,000 individual patterns on to a chip. I was able to see from the results that after the initial detection of the easy faults, only 1 pattern out of every couple hundred randomly generated would be effective, and end up detecting up to 50 hard faults. I concluded that we need either more specialized pattern templates or perhaps even to hardcode a couple of patterns to detect the hardest faults.
After double checking my data for any errors, we found a strange discrepancy in the file Fastscan gave us. According to Fastscan's output file, it detects certain faults multiple times with different patterns, which would mean that it was not "dropping faults" from the list of faults. However, if it were not dropping faults, that would mean every pattern would at minimum have to detect 1 fault every time on its output. But the file Fastscan gave us had patterns that detected 0 faults, which can't happen. We are looking into this strange error now. In addition, I found another error in the fault dictionary generator, however not one that affects any data.