Maybe I'm trying to get too fancy and find the machine state changes rather than just calculating the duration between records instead. I thought it would be computationally easier but the logic is getting quite complex.