Try some worr.. oops ferrite beads on all wires, make sure all chips are decupled with 10-100 nF caps, gove the '51 a tantalum 4.7-22 uF as well, it gives quite a kick when you write a '1' to a port pin held low. You can protect in software from sensor rattle (e. g. insist on several identical readings before accept) but if electromagnetic noise enters your board, you have had it. The issue is 1) if your program goes astray in the field, but not in the lab lt's EMF. If your program goes astray in the lab as well it is either shitty hardware design or sensor rattle.

Have fun

Erik