I haven't used any Python TDMS libraries, but a quick Google search turns up several, so that might be worth a try.
Perhaps you should start by logging all your data to a text file, rather than binary, so you could open it in Excel or a text editor and see whether you're even capturing the data you want. For example, the "Elapsed Time" function isn't doing what you want, because it resets after 100000 seconds. Instead, you should get the date and time before the while loop starts, then get it again after the while loop ends (one easy way to do this is to put a Get Date/Time in Seconds function inside the while loop, wired to the while loop border, so you get the value from the last iteration of the while loop) and subtract to get the elapsed time.
There is no need for a sequence structure in your VI.
There is no such thing as "columns," or for that matter an array, in a binary file. It all depends on how you interpret the data when you read it.
Instead of reading the data as a waveform, read it as a 2D array of DBL. You could simply write that array directly to the binary file, without any of the dynamic data and waveform manipulation. What you'll get in the file is 512 samples of the first channel, followed by 512 samples of the second channel, repeating. You could transpose the array first, so that the data would alternate between samples of the two channels, on sample at a time (instead of in blocks of 512).
