In a traditional fault detection
and diagnosis (FDD) scheme, all
of variables in a multivariate system are examined together as a single
group. This FDD scheme may suffer from the amplification and masking
effects that are caused by introducing fault-free variables in the
fault detection index, especially for a multivariate system with a
large number of variables. To overcome this problem, a new FDD scheme
for multivariate systems is developed using variable grouping. First,
a data-driven variable grouping algorithm is proposed to divide system
variables into groups. The optimal variable grouping is obtained by
maximizing variable correlations within groups but minimizing variable
correlations among groups. Then, a multigroup FDD scheme is developed,
consisting of an intergroup FDD method and three intragroup FDD methods,
called the group-based T2 method, the
dominative latent variable method, and the intragroup regression method,
respectively. Because each group consists of a few closely correlated
variables, the fault detection index in every group can reduce the
amplification and masking effects caused by redundant variables. This
makes the fault detection index more sensitive to faults occurred
in variable groups, and, hence, fault detection performance is improved.
Moreover, the multigroup FDD scheme can clearly reveal the faulty
groups in which faults occurred and the major contributing variables
in the faulty groups. This facilitates the diagnosis of faults. The
advantages of the multigroup FDD scheme are demonstrated with two
case studies.