We run the plasma process fully automated with robots, some with cold wire feed, some with hot wire feed. Your process sounds like a cold wire feed system so 2 major factors come to mind. 1: The position of the wire feeding into the puddle must be perfectly aligned to the joint in the direction of travel on the front edge of the plasma arc. The wire should enter the plasma column .5-.75 mm above the plate being welded to ensure proper melt-off without making the plasma column split or deflect. Too high, the wire will glob up and deflect the plasma column. Too low and it will curl up off the plate before it makes it's way into the plasma column, causing "shark tooth" points on the toe's of the weld joint. 2: Plasma tip orifice could be too large for the tungsten being used making the gas column too wide and not focused. If it's too small it will not let gas flow fast enough to ionize at the correct rate. This could cause the arc to flow to one side or another, even circular wandering is possible.
One other thing is to keep the tungsten ground with the same angle and depth of recess into the tip. If the depth of recess is not the same, major gas flow changes will affect arc properties. E-mail some pictures of the off location weld, or even your set-up prior to weld and maybe we can get to the bottom of the problem: mdgwelding@yahoo.com

Mark