I am fairly certain that I understand the setup, so I'll go ahead with an explanation.
I'll start with the bad news and get that out of the way. The way your experiment is constructed is an extremely complex design structure. In addition, the three sensors and the split batches are neither true repeats nor true replicates. The three sensors have sensor to sensor differences as well as positional differences included while the split batches have time differences and non-setup related variations. Normally, this type of experiment would be setup as a split plot or a least a blocked design. However, it is too late to do so now.
Now the good news: Both the split plot and blocked designs are intended to deal with large sources of noise variation and situations where the level of noise varies from whole plot to subplot. In your experiment, the variation from sensor to sensor, and from half batch to half batch is quite small. I replaced your repeat column with two columns; one for sensor location, the other for half batch. These were not even close to significance. While the following is technically a cheat, it is a very practical solution to your issue.
I would treat these as if they were true repeats. That is, set up six columns for the six results. Take the mean of the six values and put it in a seventh column then analyze that column as your response. If minimizing variation within a run is important, you can take it a step further and put the standard deviation of the six columns in an eighth column and analyze that in addition to the mean.