An adaptive pseudo-inverse method for the fault-tolerant output allocation in linear observers with redundant sensors