Better justification for splitting
I have discovered that for the same concentration, there are multiple rows. After analyzing these rows, I noticed that the weld ID has a different suffix depending on the experiment. Therefore, it makes perfect sense to have complementary null values, since each experiment does not measure the same features.