Spectral conditioning within the energy graph-based visualisation fault detection method

Abstract

In an effort to improve fault detection performance, the potential of the spectral theorem as a conditioning technique is explored within an energy graph visualisation (EGBV) fault detection method. Two conditioning techniques consistent with the spectral theorem, namely, sample self-comparison and hybrid sample self-comparison, are proposed as modifications of EGBV, followed by comparison with the standard EGBV algorithm using fault data from a continuous stirred tank reactor pilot plant. The results suggest that the former and latter provide better performance compared to the standard EGBV algorithm.