(268e) Fault Detection of Non-Gaussian Processes Based On Reconstruction

In this paper, a fault reconstruction algorithm based
on fault-relevant-directional Kernel ICA is proposed. Compared
with the traditional fault reconstruction method whose fault
model is composed of the first major distribution directions,
the proposed reconstruction algorithm gives a deep analysis
of the original fault space according to the relationships with
normal process information to extract the principal directions
that are relevant to, or affected by fault. New reconstruction-based modeling and monitoring approaches are proposed.
The proposed method is applied to process monitoring of the
operational process of the EFMF. Application results indicate
that the fault-relevant directions are effectively extracted by
using the proposed approach to build the fault-relevant process
model and improves the fault identification ability. The motivation of this paper is to
optimize the model used for fault reconstruction in feature space.
Currently, most research work reported in the literatures for
fault reconstruction does not consider the relationship between
normal status and fault case. In many practical applications, this
relationship as well as the nonlinearity of industrial process has
to be considered. Since this problem is hard to solve by using
traditional methods, a new fault reconstruction method based on
fault-relevant-directional Kernel ICA is proposed.