Enhanced auto-associative neural networks for sensor diagnostics (E-AANN)

We address the problem of sensor fault diagnosis in complex systems. The motivation for this work is the common problem encountered in industrial setting, i.e. sensor shift, drift and outright failure. The approach proposed in this paper is based on auto-associative neural networks but has been extended to address some intrinsic deficiencies of these types of networks in practical setting. In particular, it is shown that the proposed approach provides the basic functionality needed for single sensor fault detection in a multi-sensor environment with limited additional computational burden. This work is presently under further development to address multi-sensor failures.