A multivariate self-tuning controller for run-to-run process control under shift and trend disturbances

Research output: Contribution to journalArticlepeer-review

40 Scopus citations

Abstract

Many manufacturing systems are controlled with PID-type controllers. In some industries, such as semiconductor manufacturing, specifications or changing conditions impose a need for adjusting such controllers on a run-to-run basis. This need has originated a collection of techniques called run-to-run process control. Self-tuning control, where model estimation is done on-line, has been shown to be a feasible tool for run-to-run control in single input-single output systems. This paper presents a self-tuning multiple input-multiple output controller for run to run control. The controller compensates for a variety of disturbances frequently found in semiconductor manufacturing, such as offsets or shifts and trends. The controller also compensates for autocorrelated responses or noise terms, for coupled responses, and for non-square systems (i.e., the number of inputs may be greater than the number of outputs). A sensitivity analysis is presented to show the performance of the controller under various system-noise combinations. A performance measure is developed to compare the behavior of the controller with that of a minimum variance multivariate controller. A multivariate EWMA monitoring chart is added to the controller as a deadband in order to trade-off the number of recipe (setpoint) changes against the variance of the outputs. This approach is contrasted with the classical strategy of trading-off input versus output variances.

Original languageEnglish (US)
Pages (from-to)1011-1021
Number of pages11
JournalIIE Transactions (Institute of Industrial Engineers)
Volume28
Issue number12
DOIs
StatePublished - 1996

All Science Journal Classification (ASJC) codes

  • Industrial and Manufacturing Engineering

Fingerprint

Dive into the research topics of 'A multivariate self-tuning controller for run-to-run process control under shift and trend disturbances'. Together they form a unique fingerprint.

Cite this