I suspect the idea is to have them longitudinal to the bar, spaced 120 degrees apart around it, so that
no matter what the bar's rotation is they can measure the amount of flexing.
If you measure the no-load value for each one, then when you measure the differences under load for
each of the three you can turn that into a quadrature value using the Clark transform Alpha–beta transformation - Wikipedia
and its easy to compute the amplitude of a quadrature signal using Pythagorus.