The importance of gross error detection and identification is widely acknowledged, but so far relatively few publications in the chemical engineering literature have addressed this problem for stochastic dynamic systems. Also, statistical process control (SPC) chart techniques have been developed and used extensively to detect shifts (a change in steady state) in the process means but have been hitherto ignored in applications involving detection of gross errors in chemical process data. In this paper a composite test procedure for detecting and identifying gross errors is proposed: we first make use of the CUSUM test (which underlies the CUSUM chart used in quality control applications) to detect the presence of gross errors. We then apply the generalized likelihood ratio (GLR) method to identify as well as to estimate the magnitudes of the gross errors. We concentrate our treatment on discrete-time, linear dynamic systems operated around a nominal steady state and corrupted by process and measurement noises, which are assumed to be Gaussian. These noises can be white or serially correlated. Computer simulation studies and some analytical results are presented.
ASJC Scopus subject areas
- Chemical Engineering(all)
- Industrial and Manufacturing Engineering