Performance assessment measures for univariate feedback control

Abstract

A normalized performance index, η(b), is introduced to characterize the performance of feedback control schemes. η(b) is attractive because it provides a measure of the proximity of control to minimum variance control, which is the optimal feedback control provided that the process can be described by a linear transfer function with additive disturbance. Both time domain and spectral interpretations of this performance index are discussed. A fast, simple, on‐line method for estimating η(b) is given, along with some of the statistical properties of the estimator. Simulation and industrial examples demonstrate the utility of η(b/)

Closed-loop identification with a quantizer

“…As was shown in [24,25], y includes an irreducible minimum variance component which remains consistent for different controllers. The standard deviation of the minimum variance component of the process output (r mv ) can be used to give an estimate of r a for specifying a suitable quantizer interval for CLID.…”

Detection of multiple oscillations in control loops

“…The first author together with staff of the Eastman Chemical Company identified tag 22 as the root cause of the oscillation at 337 samples per cycle using minimum variance controller performance benchmarking [21], op-pv maps (plots of controller output versus process variable), and non-linearity analysis [15].…”

The impact of compression on data-driven process analyses

“…The widely used Harris index [26] is a minimum variance benchmark of control loop performance. Significant industrial implementations are being reported [27][28][29].…”

“…The Harris indexes for the three data sets were calculated using the method described in Desborough and Harris [26] with an estimated time delay of five samples. The index is determined from the residuals between the measured controller error denoted by y and a b-step ahead prediction,ỹ y.…”